I.           Introduction

Artificial intelligence (AI) is being adopted rapidly across the Philippines—in government services, banking and finance, agriculture, education, health, media, labor platforms, the creative industries, and law enforcement—often without adequate public consultation, regulatory oversight, or safeguards, especially for vulnerable communities. While AI presents opportunities for development and efficiency, it also carries significant risks: algorithmic discrimination, labor displacement, environmental degradation, surveillance, disinformation, data privacy violations, and the erosion of rights for already-marginalized groups, including informal workers, indigenous peoples, women, LGBTQIA+ people, persons with disabilities, and rural communities.

Civil society organizations (CSOs) and grassroots groups are often the first to witness and respond to these impacts on the ground, yet are frequently excluded from national conversations on AI governance, which tend to be dominated by industry and technical experts. This report consolidates community-level knowledge and lived experience into a coherent, evidence-based risk map intended to inform advocacy and policy engagement on AI in the Philippines.

Objectives

This report aims to:

• Document and validate observed impacts of AI within specific sectors and communities in the Philippines;
• Map key risk areas—social, economic, political, environmental, and rights-based—associated with AI deployment in the Philippine context;
• Identify priority issues and affected groups requiring urgent policy attention;
• Set out initial recommendations to inform policy engagement; and
• Support a shared network and community of practice among CSOs working on AI and digital rights in the Philippines.

Scope and contributors

This report draws from feedback solicited from risk-mapping workshops organized by the Foundation for Media Alternatives and attended by civil society organizations working on human rights, labor, consumer protection, women’s rights, LGBTQIA+ rights, indigenous peoples’ rights, media and press freedom, and environmental and climate advocacy; grassroots and community-based organizations; sectoral representatives from informal work, farming, gig/platform work, and urban poor communities; and academics, technologists, journalists, and legal practitioners in digital rights.

This report also utilized feedback from practitioners in public health and health communications (including tobacco and drug-control policy and adolescent health), a university data-protection office, climate and environmental research and policy institutes, organizations working with mining-affected and indigenous communities, and workers and artists in film production, illustration and graphic design, marketing, theatre, and community arts.

Methodology

• Workshop 1 (28 July 2026, Quezon City) brought together participants from sectors including labor and business process outsourcing (BPO), education, law, human rights, and media, through facilitated sectoral discussions and small-group risk-identification exercises.
• Workshop 2 (23 September 2026, Quezon City) responded to gaps in the first workshop, which had no environmental representation and little input from the health and creative sectors. It opened with a short plenary presentation of an AI risk ladder as a shared vocabulary, after which participants split into four focus group discussions (FGDs): health, women and gender, environment, and the creative sector.

In both workshops, discussion was structured around guide questions: where and how AI is having an impact within a given sector; who benefits and who is harmed or excluded; what concrete incidents or patterns have been observed; and what existing laws or protections, if any, apply. Risks were plotted on a likelihood/severity matrix, and priority themes used as the basis for policy recommendations addressed to government, regulators, companies, and AI developers.

I. Overview of proposed AI regulations in the Philippines

As of date, seven bills on artificial intelligence were filed in the 20th Congress of the Philippines between June and November 2025. These bills have been referred to a Technical Working Group (TWG) of the House Committee on Information and Communications Technology. The TWG is currently consolidating these draft policies into a substitute bill, the proposed Artificial Intelligence Development and Regulation Act.[1]

The bills filed so far take three broad approaches. Five bills propose a general regulatory framework with a new oversight body: HB 13 (an AI Board under the DICT), HB 57 (a Bureau of AI Systems under the DICT), HB 73 (an AI Development Authority under the Office of the President), HB 3462 (a Philippine AI Commission) and HB 6028 (a National AI Commission chaired by the DOST Secretary). Each bill differs in scope and the extent it delegates powers to the created agencies.

HB57 and HB13 define rights, prohibited practices and penalties in the bill text itself, while HB 73 and HB 3462 leave most of the substantive rules to the regulator. HB 2323 takes a planning and coordination approach, and creates an AI Commission under the Office of the President to adopt a national strategy and action plan, with a ₱50 million initial appropriation.

Distinctly, HB3905 takes a penal approach, treating the use of AI in a crime as an aggravating circumstance rather than regulating AI systems.

The bills also differ on penalties and funding. Maximum criminal penalties of imprisonment range from six years (HB 13, HB 57) to twelve years (HB 6028). HB 3462 provides administrative fines only, and HB73 and HB2323 set no penalties of their own. HB 2323 is the only bill that appropriates a specific amount to fund implementation.

The table below summarizes key provisions on each bill, based on nine specific criteria set out by the TWG. “Yes” means the bill itself contains the provision; “partial” means it is limited, delegated to a future regulator, or stated only as a principle. The section numbers / provisions relied on are shown in parentheses.

 

Criterion	HB 13	HB 6028	HB 2323	HB 3462	HB 73	HB 57	HB 3905
Risk-based classification	No	Yes, three tiers (16)	No	Yes, two tiers (3, 5)	No; left to regulator	Yes (15)	No
Prohibited AI practices	Yes (24)	Partial; offenses, no banned uses (20)	No	No	No	Yes (20, 32)	No
Transparency / disclosure	Yes (5(d))	Yes (17, 20(d))	No	Partial; high-risk only (7)	Partial; delegated (7(a))	Yes (17, 19)	No
Individual rights	Yes (5)	Partial; redress channel (12)	Policy only (2)	Principles only (7)	Partial; complaints mechanism (7(h))	Yes (5, 16, 34)	No
Enforcement powers	Yes; AIB (9)	Yes; NAIC (9)	No	Yes; PAIC (5, 9)	Yes; AIDA (7(d))	Yes; BAIS (7)	Existing courts; DOJ rules (6)
AI registry	Partial; companies, not systems (23)	Yes; all systems (11–13)	No	No	Partial; registration delegated (7(e))	Yes (7(c), 15)	No
Sector-specific rules	Partial; labor (19-20)	Partial; labor (18)	Partial; agency roles (13)	No	No	Yes (21–28)	No
Specific appropriation	No; GAA only (27)	No; DOST budget (21)	Yes; ₱50M (19)	No	No amount (10)	No; DICT budget (76)	No
Max penalty	6 mo–6 yrs prison; no fine (25)	₱10M and/or 6–12 yrs (20(b))	None	₱10M administrative fine (9)	None; existing laws (11)	₱10M and/or 6 mo–6 yrs (32)	+1 degree; +2 for enabling entities (3, 4)

 

The assessments below discuss the criterion as applied to each bill in detail.

House Bill 13, Artificial Intelligence Development and Regulation Act  

HB13, the proposed Artificial Intelligence Development and Regulation Act, provides for a rights charter, creates three new government bodies, and adopts a national AI strategy.

Sec. 6 creates the Philippine Council on Artificial Intelligence (PCAI) under the DICT as a policy-making and advisory body of experts, which includes an ethicist, a human rights advocate, and a citizens’ representative. Its functions under Sec. 7 include issuing a governance framework, as well as developing a code of ethics for developers, rules against harmful applications, and an AI risk reduction plan. Sec. 9 creates a separate regulator, the Artificial Intelligence Board (AIB), chaired by the DICT Secretary with the Secretaries of DOST and DTI and the heads of IPOPHL, the National Privacy Commission and NICA. The AIB is given substantial enforcement tools: it may investigate on its own initiative or on complaint, impose fines, issue subpoenas, inspect premises, and appoint automation auditors to review an algorithm’s behavior. Sec. 11 creates a National Center for AI Research (NCAIR) to lead research and skills development.

The bill features an AI Bill of Rights in Sec. 5. It sets out five rights: protection from unsafe and ineffective systems, with pre-deployment testing and ongoing monitoring; protection from algorithmic discrimination, backed by equity assessments, disparity testing and public algorithmic impact assessments “whenever possible”; privacy by design consistent with the Data Privacy Act; a right to know, requiring plain-language notice and explanation of outcomes; and a right to remedy, including opting out in favor of a human alternative and contesting AI’s impacts.

Notably, the bill also addresses labor. Sec. 19 allows employers to adopt AI only if they issue an AI Governance Policy explaining its use and data handling. Sec. 20 entitles workers made redundant by AI to separation pay of at least one month’s pay or one month for every year of service, whichever is higher, and directs the DOLE to create a Job Displacement Program.

Sec. 24 prohibits lethal autonomous weapons; AI that manipulates a person beyond their consciousness in ways likely to cause physical or psychological harm; and AI-based social scoring. Sec. 26 makes directors and officers of a juridical entity personally liable via imprisonment.

However, other provisions of HB13 may raise concerns. The bill has no risk classification: obligations do not vary with how risky a system is, and the Bill of Rights in Sec. 5 is framed as rights by which development “shall be guided” rather than as duties with specific penalties for non-accomplishment. Sec. 24 makes it unlawful to use AI that causes “unnecessary, unjustifiable and indiscriminate moral or pecuniary damage”, a standard broad enough to raise vagueness concerns, especially since Sec. 25 attaches a penalty of six months to six years’ imprisonment to such offense.

Sec. 22 limits public disclosure of trade secrets and proprietary data. Further, the central database provided in Sec. 23 lists AI companies, their trademarks and their violations rather than individual AI systems. It is also concerning how Sec. 3(l) defines social scoring narrowly as rating influence on social networks, which may leave out scoring used for credit, employment, or public services.

Sec. 27 makes no specific appropriation for implementation, leaving funding to the annual General Appropriations Act.

HB 6028, Artificial Intelligence Regulation Act

HB 6028, the proposed Artificial Intelligence Regulation Act, introduces a registration-based regime administered by a new National AI Commission (NAIC). The Commission is chaired by the Secretary of Science and Technology and includes agency representatives alongside private sector and civil society members (Sec. 7). Its powers under Sec. 9 include maintaining a national registry, classifying systems by risk, certifying and auditing applications, suspending non-compliant systems, hearing complaints, and imposing administrative penalties. The bill also creates an AI Ethics Review Board (Sec. 15) and requires a multi-year AI roadmap from the NAIC (Sec. 10).

The creation of an AI registry is the bill’s most notable feature. Sec. 11 requires anyone who develops, deploys, operates, imports, sells or provides access to an AI system in the Philippines to register it with the NAIC, and Sec. 13 makes registration a precondition for deployment. Sec. 12 specifies what a registration must contain, including the system’s intended purpose and sector, its foundation model, its data sources and safeguards, its risk classification, and a mechanism for redress, among others. The registry, combined with a public online portal for complaints and harm reports under the NAIC’s mandate, gives regulators an inventory of deployed systems and gives the public a channel for redress.

Further, Sec. 16 classifies systems into three tiers based on risk levels. A high-risk system is one whose decisions significantly affect public safety, identity, fundamental rights, livelihoods or access to essential services. These systems include those used for education, healthcare, law enforcement, critical infrastructure, justice, finance, and public administration. Moderate-risk systems have limited or reversible effects, and do not directly compromise safety, rights, or critical services. These include systems used in process automation, service optimization, internal evaluations, and resource allocation. Meanwhile, low-risk systems have minimal impact potential, and refers to those used for non-critical tasks such as administrative support, personalized content delivery, or basic data visualization, which do not materially affect rights, safety, or public interest.

Sec. 17 places concrete duties on developers and deployers, among them disclosing to users when they are interacting with AI; conducting algorithmic audits; and accepting liability for harms and maintaining a mechanism to shut a system down.

Sec. 18 requires an AI Deployment Impact Report sixty days before AI is introduced in ways that affect workers. It also requires mandatory retraining and transition assistance, and a tripartite oversight council to monitor AI-induced layoffs.

However, other provisions of this bill may require further consideration and consultation. Sec. 11 requires every AI system to be registered, whatever its risk tier. This is misaligned with the three-tier classification in Sec. 16, and will mandate low-risk systems to carry the same registration burdens as high-risk ones. The penalty scheme in Sec. 20 sets out seven offences, each punishable by a fine, imprisonment, or both. The heaviest, using AI to commit fraud, crime or harm, carries a penalty of ₱2 million to ₱10 million or six to twelve years; arguably, this overlaps with offenses already defined in existing penal laws. The offense covering AI-enabled disinformation and surveillance carries a penalty of up to to ten years, and its reach will depend heavily on how those offenses are defined.

Sec. 21 provides no specific appropriation for implementation. Initial costs are charged to the current yearly budget of the Department of Science and Technology, with later funding left to the annual General Appropriations Act.

HB 2323, Artificial Intelligence Act of 2025

HB2323, the proposed Artificial Intelligence Act of 2025, is a governance and development bill rather than a regulatory bill. Its aim, as stated in its title, is to mainstream AI into government policy and to establish a national framework strategy. Sec. 4 creates an AI Commission attached to the Office of the President as the sole policy-making body for AI. Under Sec. 5, the President chairs the AI Commission, with three appointed Commissioners and a twenty-member Advisory Board drawn from government, local government leagues, and the private sector.

Planning provisions feature dominantly in the bill. In Section 10 to 11, the Commission must adopt a Framework Strategy and Program on AI within six months of effectivity, to be reviewed every three years. A National AI Action Plan must follow within a year, covering impact assessment, the identification of vulnerable populations, differential sectoral impacts and risk management (Sec. 12).

Sec. 13 assigns functions to fourteen agencies. For example, DepEd and CHED are to integrate AI and AI ethics into curricula; DICT to formulate a plan to build data centers under data sovereignty principles; DOLE and TESDA to study effects on livelihoods; DOH to study how to expedite medical service delivery and bolster equal distribution and medical services; DOST to spearhead AI promotion in academic institutions; and the Commission on Human Rights to examine AI bias and discrimination. Sec. 14 tasks the Commission to coordinate with NGOs and CSOs, academe, people’s groups, private sector and corporate groups, and other stakeholders in developing and implementing the National AI Action Plan and action plans at the local level.

The bill also details funding and coordination. Sec. 19 appropriates ₱50 million as an initial operating fund from the President’s contingent fund, with later funding through the General Appropriations Act, while Sec. 16 directs agencies and local governments to budget for AI programs. Accountability to Congress is operationalized through an annual report (Sec. 17) and a Joint Congressional Oversight Committee (Sec. 18).

The bill’s weaknesses follow from its limited scope. It does not contain prohibited acts or penalty provisions. It also fails to establish a registry or a formal risk classification. Rights appear only in the Declaration of Policy (Sec. 2), which refers to privacy of communication and preventing discriminatory systems, without enforceable duties attached. While Sec. 4 describes the Commission as independent, Sec. 5 makes the President its chair, and Sec. 20 assigns the promulgation of implementing rules to the DOST rather than to the Commission itself.

HB 3462, Philippine Artificial Intelligence Act of 2025

HB 3462, the proposed Philippine Artificial Intelligence Act of 2025, is a short bill that creates a Philippine Artificial Intelligence Commission (PAIC) and sets baseline duties for high-risk systems. Sec. 3 defines a high-risk AI system as one that poses a significant risk of harm to health, safety or fundamental rights. Under Sec. 5, the PAIC has several responsibilities: to classify systems as high- or low-risk; to conduct audits; to issue ethical guidelines; and to impose penalties.

Substantive provisions for AI developers and users are found in Sec. 7 and 8. Sec. 7 adopts four principles they should adhere to: human oversight and control, privacy by design consistent with existing data privacy law, algorithmic transparency for high-risk systems, and freedom from discriminatory bias. Sec. 8 holds developers and users responsible for harm caused by high-risk systems unless the harm results from the user’s gross negligence, and requires comprehensive audit trails.

Sec. 8 is significant for imposing simple accountability rules in case of harm. Placing responsibility for harm on developers and users of high-risk systems, with gross negligence by the user as the only stated defense, gives affected persons a clearer route to remedy than a fault-based rule. Anchoring the definition of high risk on fundamental rights, and requiring audit trails, also supports later review of contested decisions.

However, HB 3462 does not establish a registry nor creates sector-specific rules. It also does not have an appropriations clause. Its classification of AI systems only has two tiers, and apart from transparency and audit duties, a high-risk designation attracts no other duties. The penalties under Sec. 9 do not correspond to a specific definition of prohibited acts, and instead apply vaguely to “violations of this Act or the rules and regulations issued by the Commission”. Nevertheless, such broad violations are punished only by fines of ₱100,000 to ₱10 million depending on severity, and the suspension or revocation of permits, licenses or government contracts. The same section allows “other such penalties as may be prescribed by the Commission,” which leaves the range of sanctions open to agency discretion, and may face challenge as an undue delegation of legislative power.

HB 73, Artificial Intelligence Development and Regulation Act of the Philippines

HB73 is a framework bill built around the Artificial Intelligence Development Authority (AIDA), a new agency created in Sec. 5, under the administrative supervision of the Office of the President. A Board of Trustees governs the AIDA, and it is chaired by the head of the DOST Information and Communications Technology Office, with private sector and academic experts as permanent members, and the heads of agencies such as the National Privacy Commission, the DOJ Office of Cybercrime, the PNP Anti-Cybercrime Group and the SEC as ex-officio members. A Secretariat headed by an Executive Director enforces the Act (Sec. 6).

Sec. 7 lists AIDA’s regulatory powers. AIDA is tasked to develop a national strategy that includes requirements for AI impact assessments, data protection and transparency in AI decision-making (Sec. 7(a)); set standards for legal, ethical and responsible AI and enforce them through investigations and penalties (Sec. 7(d)); and establish licensing and certification requirements under which developers and deployers must register with AIDA (Sec. 7(e)). It must also ensure compliance with the Data Privacy Act (Sec. 7(f)), set cybersecurity standards (Sec. 7(g)), create a complaints and grievance mechanism (Sec. 7(h)) and report regularly to the public on AI’s impact on society (Sec. 7(i)). Sec. 8 adds a promotional mandate, including research funding, incentives for responsible developers and talent development.

The bill’s strengths are its multi-agency, institutional design and its breadth of mandate. Bringing privacy, cybercrime, intelligence, telecommunications and securities regulators into one board gives AIDA a coordinated view across agencies. The registration duty in Sec. 7(e), the complaints mechanism in Sec. 7(h) and the public reporting duty in Sec. 7(i), all together, provide basic tools of transparency, oversight, and redress. Combining promotion and regulation in one agency also reflects the bill’s stated aim of balancing development with safeguards.

However, its main weakness is that most substantive rules are delegated rather than written into the bill. The bill does not have risk classification for AI systems, does not list prohibited acts, and imposes no transparency or explanation duty directly on developers—these are left to the strategy and standards that the AIDA will issue. The explanatory note describes required risk assessments and meaningful explanations of AI decisions, but the operative sections do not themselves require them.

Sec. 11 provides no specific penalties: violators are to be punished “in accordance with the penalties provided for under existing laws and regulations”, while Sec. 7(e) refers only to unspecified administrative sanctions. Sec. 10 states that the government “shall allocate funds” but does not appropriate any amount or name a source of funding. There are also drafting errors: Sec. 7(e) refers to the guidelines “of the Council”, a body the bill does not create, and Sec. 5 names the DOST-ICTO as chair although that office was abolished and transferred to the DICT in 2016.[2]

HB 57, National Artificial Intelligence Code of the Philippines

HB57 is a comprehensive bill, notable for having 84 sections in ten chapters. Among others, it applies to all public and private entities and to AI systems that affect Filipinos through cross-border platforms (Sec. 3). Sec. 6 creates a Bureau of Artificial Intelligence Systems (BAIS) within the DICT as a national regulatory, investigatory and technical body, which has the following powers: classifying systems by risk, maintaining a national registry, conducting pre-deployment reviews and audits of high-risk systems, suspending or prohibiting systems that threaten safety or rights, and referring violations to the DOJ, among others. Sec. 12 requires public agencies and regulated private entities that use AI to set up Internal AI Compliance Units.

The bill is notable for defining rights and duties explicitly in the text, instead of leaving them to later regulation to a government agency. Sec. 5 sets out a Bill of Rights in AI with eight rights, including meaningful human oversight of consequential decisions; algorithmic transparency; data sovereignty; contestability and redress; “non-obsolescence” of labor through just transition support for displaced workers; and citizen participation on the development of AI policies.

High-risk systems are defined and regulated. Sec. 15 lists the factors that make a system high-risk, including its effects on fundamental rights, use in critical sectors and use for biometric identification, or predictive policing. High-risk systems in the bill require pre-deployment review, mandatory registration, risk mitigation and continuous audit.

Under the bill, deployers must also disclose AI use, its purpose, and appeal channels (Sec. 17). Affected persons may demand human review of AI decisions (Sec. 16), and high-risk systems must document their training data, bias testing and decision pathways (Sec. 19). Sec. 20 prohibits generalized social scoring, deceptive impersonation or manipulation, autonomous systems without effective human intervention such as lethal force in law enforcement, and the exploitation of vulnerable groups.

Sector-specific chapters are a key defining feature of the bill. Secs. 21 to 28 set tailored obligations for government services, education, healthcare, finance, labor and employment, law enforcement, critical infrastructure, and media. Examples include Department of Health certification of diagnostic AI (Sec. 23), human appeal of AI-based credit denials (Sec. 24) and a ban on indiscriminate real-time mass surveillance without judicial oversight (Sec. 26(b)).

The imposition of penalties depends on the nature and gravity of specific violations. Unauthorized high-risk deployment carries an administrative fine of up to ₱5 million per violation (Sec. 30) while failure to disclose AI use up to ₱2 million (Sec. 31), for example. Sec. 33 makes the entity, the responsible officer and, on gross negligence, the compliance officer jointly liable for specific offenses. Affected persons may file for injunction and damages (Sec. 34), and whistleblowers are protected (Sec. 36). The bill also creates a nine-member National Panel of Experts with seats for labor, human rights, youth, regional and media representatives (Secs. 71 and 73), and the BAIS must explain in writing when it departs from the Panel’s advice (Sec. 72(e)).

However, the extent of power given to the BAIS appear problematic given its creation under the DICT. The BAIS is a bureau under the “direct supervision and control” of the DICT Secretary, and all its actions are subject to DICT policy guidance (Sec. 6), which limits its independence from an agency that also promotes AI adoption in Chapter IV (Sec. 37 to 43). Sec. 9 makes the Bureau’s interpretations of the Code binding on all government offices and private entities unless reversed by the Supreme Court, and presumes them valid; this goes further than the deference courts ordinarily give agencies.

Other questionable provisions also exist in the text. The prohibited practices in Sec. 20 and the criminal offenses in Sec. 32 do not appear to match. Exploiting vulnerable groups and deploying autonomous systems without human control are prohibited in Sec. 20 but carry no stated penalty in Sec. 32, while Sec. 32 adds biometric surveillance not listed in Sec. 20. Offences such as “deceptive political or financial manipulation” (Sec. 32(d)) are not defined, which raises vagueness concerns for a criminal provision. Further, Sec. 7(c) refers to mandatory registration by “all deployers” while Sec. 15 mandates registration only for high-risk systems, leaving the scope of the registry unclear.

Finally, Sec. 76 provides no specific appropriation: initial costs are charged to the DICT’s current budget, and the AI Safety Research Fund in Sec. 43 depends on future General Appropriations Acts.

HB 3905, AI-Aided Criminal Liability Act

HB3905, as distinguished from other pending bills, is a penal measure. It does not regulate the development or deployment of AI; it changes how crimes committed with AI are punished. The bill seeks to closes a gap that its own explanatory note identifies: existing laws such as the Cybercrime Prevention Act do not treat the use of AI or synthetic media as an aggravating circumstance.

Sec. 3 provides that any crime under the Revised Penal Code or special laws committed “by, through and with the use of” AI, machine learning, large language models or generative technologies is aggravated, and thus the penalty is raised to one degree. Use of AI in connection with rebellion, insurrection or attempted coup d’état is instead absorbed into those crimes.

Sec. 4 makes corporations, organizations and platforms that “knowingly or negligently enable” their AI to be used for crimes civilly or criminally liable, with a penalty two degrees higher. If the entity or person is foreign, its foreign stockholders are to be deported immediately. Sec. 4 thus extends responsibility upstream to developers and platforms, not only end-users.

Sec. 6 assigns the implementing rules to the Department of Justice in consultation with DICT.

Several drafting problems in the bill need resolution. Sec. 3 applies the aggravating circumstance only when AI use is “inherent in the commission” of the crime, but how to determine this is unclear. The explanatory note says higher penalties under special laws will prevail and that an additional penalty applies only where penalties are equal, but the bill itself is silent on this regard. Sec. 4 imposes a penalty two degrees higher (harsher than for persons who directly commit the crime) on entities that are merely negligent, without stating the base penalty it is measured from or how criminal liability attaches to a juridical person. The deportation clause applies to foreign stockholders without any hearing or link to their own conduct, which raises due process and equal protection questions.

III. Sectoral risk map

This section, using the methodology described in the Introduction, sets out the risks identified across sectors, organized thematically. For each theme, findings are presented in terms of affected persons, specific case studies, and general observations on AI use, impact, and risks per sector.

Statements presented in this section represent only the views of participants to the workshops organized by the Foundation for Media Alternatives, and are not meant to represent organizational or institutional view.

Labor and economic risks 

Who are affected? BPO/IT-sector workers, gig and platform workers, small and medium enterprises pushed to adopt AI without support, freelance and project-based creative workers, below-the-line film crew, community and regional artists, and workers in health communications and marketing.

BPO layoffs linked to AI adoptionAt a BPO operation in Clark, Pampanga, a number of quality-assurance specialists lost their jobs due to automation. Separately, an estimated 1,500 employees were terminated at another BPO company, reportedly connected to AI-driven survey/monitoring functions. Workers have also reported headcount reductions tied to the introduction of AI support tools within specific accounts. Content moderation and machine learning at a major platform Workers in content moderation roles at a major technology platform report risk to job security, since the moderation work they perform is itself used as machine-learning training data — raising the concern that the AI systems being trained by these workers may eventually replace them.

In the Philippines, job displacement and labor intensification (i.e., workers doing more, faster, without added compensation) are the most prevalent, cross-cutting risks mapped in terms of AI use and adoption. These two can occur simultaneously. On one hand, AI agents are trained to replace human agents in the BPO industry, with management, in some cases, raising quotas in anticipation of automation, alongside reported layoffs in quality-assurance roles. On the other hand, while workers may not always be removed from their jobs, their workload and stress may increase: tools that seek to facilitate efficiency in work may simply raise expectations and deadlines, and make work environments more demanding.

This extends to the creative and communications professions. In the film industry, the pace by which AI is adopted was described as exponential in scale and speed, outstripping institutions’ capacity to train workers.  Film editors now face autonomous AI-driven editing. Below-the-line film crew (for example, camera assistants whose focus-pulling is now handled by AI-enabled cameras) must buy expensive tools and retrain at their own cost. This is exacerbated by the observation that the displacement of tasks is not matched by the creation of new roles or by capacity-building.

Digital artists reported situations where professionals readily use use for creative work where they would otherwise have to pay a creative worker. Freelance illustrators and designers reported losing clients and commissions, including from advocacy organizations and newsrooms that now use AI-generated images. For the digital artists, this came with the realization that some personnel in organizations and newsrooms may be comfortable using AI as long as their own (non-creative) roles are not at risk. Participants also described adoption as increasingly coerced rather than chosen. As one designer put it, “either I adapt or I lose my job”. A manifesto circulated among artists sought to preserve some autonomy for creative workers whose employers require them to use AI, but since clients were identified as the top beneficiaries of these changes, participants saw these as shifting bargaining power further away from creative workers.

Representatives from the creative sector also note how marketing teams were restructured or outsourced as AI was introduced, and how community artists in the provinces lost paid commissions for local tourism and school materials. Marketing artists working under ever-tighter deadlines added that stock photo libraries, a basic resource of their work, are now flooded with AI-generated images that only a handful of people can distinguish from human-made work. Young artists and artists from outside the city were described as facing even fewer opportunities and spaces than before. A health communications practitioner described how a company restructured into project-based roles without government benefits, as research teams pushed automation.

Transparency, bias, and technology-policy gaps

Who are affected? Essentially all users of AI-enabled services, including customers routed through AI-driven customer service, and specifically neurodivergent individuals, who are more vulnerable to forming false attachments to AI systems that simulate empathy (a dynamic sometimes termed an “intimacy trap”). Communities absent from training data (for example, people with autism, trans and gender-diverse people, and indigenous communities) are affected by misrepresentation or exclusion.

Participants described how large language models function largely as “black box” systems—their training data, biases, and decision logic are not disclosed, often on the grounds of trade secrecy. Without disclosure of training data and bias-testing methodology, affected communities cannot assess or contest AI-driven decisions. Participants described an “asymmetry” rather than simple misinformation arising from the use of data for AI: models favor data that is accessible, paid for, or already dominant, so certain sources are mainstreamed and others disappear.

Workshop participants gave some practical examples of what this means: an AI tool proposing outdated or Western-framed ideas for an autism awareness campaign, that the community concerned found obsolete; AI tools giving confident but out-of-date interpretations of health regulations; AI summaries that do not distinguish industry-funded studies from independent ones; and output quality that depends on who can pay for the better tiers or on know-how for prompting efficiently.

Risks to gender and women’s rights; technology-facilitated violence

Who are affected? Women and girls generally, with particular concern for indigenous girls, girls and LGBTQIA+ children, women human-rights defenders, and women in extractive-industry frontline communities. Ordinary survivors are affected more than high-profile ones, since participants observe that cases involving celebrities attract faster action and protection, while ordinary individuals face many barriers to access justice.

Discussions by women’s and gender rights’ groups started on how technology-facilitated gender-based violence (TFGBV) has changed from crude image manipulation to sophisticated software. A case mapping by a women’s rights organization found AI-generated sexual material uploaded to pornographic sites and circulated through messaging platforms, including secure messaging apps. Researchers working with environmental human rights defenders (EHRDs) similarly reported cases of TFGBV in which AI was used to generate images of defenders that were not their own. Participants working with indigenous women reported that dating apps and social media that match users by proximity are connecting girls of about 15 years old with adult men. In conditions of landlessness, poverty, and limited access to education, these girls leave with men they have just met, only to be beset by early pregnancy and later on, abuse and abandonment: a form of long-term TFGBV that begins online. In communities that rely on pay-per-use wifi, access is limited to brief, unsupervised sessions.

Participants also reported that online advocacy for safe abortion and for LGBTQIA+ rights is hit by automated takedowns triggered by flagged words, and that platform community guidelines are written from a Western perspective. Both of these risks are amplified by the adoption of AI. Advocates have responded with their own low-technology alternatives, such as a web-based FAQ platform that guides women seeking information on safe abortion. Women and LGBTQIA+ people were also identified as targets of AI-enabled scams.

This risk amplification underlies the issues raised by participants in the women’s and gender rights sector. Participants are concerned about a steep increase in AI-generated child sexual abuse material as documented by the Internet Watch Foundation,[3] and as reported to the House Committee on Welfare of Children in 2025, alongside continued use of deepfake and “nudify” applications. Yet existing safeguards—such as the anti-OSAEC and anti-child sexual abuse material laws, the Data Privacy Act, the Cybercrime Prevention Act, the Safe Spaces Act, and the Anti-Violence Against Women Law—are being tested by generative tools that did not exist when those laws were drafted, which makes it hard for survivors to determine the appropriate legal remedies.

Participants also noted that women’s groups have largely not been consulted on proposed AI bills that are not specifically about gender. In their view, attention to AI-enabled gender-based violence has been “sensational”: proposals to amend the Safe Spaces Act to cover AI-generated content followed deepfakes of public figures, such as the actress Angel Aquino and a social media influencer, while violence against ordinary women draws little attention until a prominent figure is involved.

In its own risk mapping, the women and gender focus group ranked AI-enabled TFGBV, including gendered disinformation, as its foremost priority, and traced it to several structural causes: AI frameworks and policies that are not gender-responsive, low AI awareness and literacy, the underrepresentation of women in AI development, the lack of platform accountability, and bias and prejudice inherent in AI models. These participants ranked second the disconnect between AI use and advocacy, as when anti-mining campaigns make use of AI whose infrastructure begins with the mining of critical minerals, and third the digitization of culture and creativity, which it framed as a tension between preservation and resistance. Both are discussed further in the sections on civic space and on the creative sector.

As an aside, participants also noted that AI can be used for good: a women’s organization’s AI-persona campaign allowed survivors to tell their stories without revealing their identities and reached a wide audience, although it drew criticism over surveillance concerns about the platform and over whether AI personas displace actors.

An AI-persona campaign for survivors of violence Two or three years ago, a civil society organization partnered with a digital advertising company’s corporate social responsibility team and the Commission on Human Rights to tell the stories of survivors of domestic violence and sexual assault through AI-generated personas, so that survivors did not need to reveal their identity. The campaign drew wide engagement and a large increase in the organization’s audience on a short-video platform, appreciation from survivors who found a safe way to share their stories and information on where to seek help, and criticism: human-rights defenders raised surveillance concerns about the platform, which participants described as owned by a Chinese company, and commenters questioned the use of AI by a digital rights advocate. Some asked whether it deprived actors of work, while others argued that it spared survivors, and actors who might have to relive trauma, from exposure.

Risks to LGBTQIA+ and gender-diverse people

Who are affected? Trans women and gender-diverse people, particularly those who have transitioned socially or medically and whose documents do not match their appearance; trans sex workers; younger trans women who obtain hormones informally.

Discussions in this sector revolved around how AI systems inherit the biases of their developers and data, and how most still follow a male-female binary. Participants described deepfake “trans-baiting” videos that stereotype and objectify trans women, a circulating AI-based checklist for identifying trans women, and reported uses of AI by anti-gender groups to attack gender-diverse people, following patterns seen in the United States. A trans-rights organization also referred to international research, warning that facial-recognition systems built on binary assumptions discriminate against trans people.

In this regard, a participant noted that, if government agencies and banks adopt AI-assisted verification for identity documents, passports, and banking, a mismatch between a person’s appearance and recorded sex can trigger suspicion of fraud, additional document requirements, or denial of entry abroad. Incidents reported include a vlogger denied entry in Dubai and a trans journalist denied entry in Macau, which a participant said arose from the absence of legal gender recognition—which AI-assisted verification makes more consequential. From this context, surveillance and profiling thus has the potential to “out” people.

Data practices such as sex-disaggregated HIV screening data count trans women with men who have sex with men, erasing their separate experience. Additionally, because the Philippines lacks a gender-affirming health policy, some trans people may turn to AI for self-medication guidance; participants described this as a risk but, if designed with human oversight and a harm-reduction approach, a possible source of support. It was also noted that AI-generated adult performers could displace trans sex workers, although this was raised only as a possibility, and is not yet reported by the community itself.

Risks in surveillance, the civic space, and human rights

Who are affected? Activists, journalists, and human rights defenders, particularly those working on politically sensitive issues; women human rights defenders; indigenous and mining-affected communities; LGBTQIA+ advocates.

Human rights defenders and activists face AI-amplified surveillance, digital harassment, and red-tagging, with open-source data (location sharing, social media posts) used to profile and target individuals. Participants note that automated content moderation takes down advocacy accounts, and that AI is used to police gender expression. Communities also described being intimidated by drones used by mining companies to assert that they hold permits. The environmental focus group specifically rated the criminalization, militarization, and surveillance of communities and environmental human rights defenders as high in both likelihood and severity.

On the other hand, AI has been noted to have protective uses in this space, such as translation, grant-writing support, and open-source intelligence for documenting human rights violations. This illustrates that the same tools carry both risk and benefit for this community. On the same note, participants also raise the idea of community-owned technology: locally controlled tools that run on local networks or offline, which would let women in affected communities share information and report violence without depending on big technology platforms. Such tools face real obstacles (cost, usability, signal, and the abandonment of earlier open-source options), but were seen as worth exploring.

AI-generated songs at a rally and in community settings An organization working with indigenous women found that songs it had enjoyed at a rally had been generated by AI, which it noticed only later from the odd pronunciation of certain words. The group saw similar uses across the city, in hotels’ welcome songs, in other indigenous communities, and in captions and posters by NGOs, and said it was now debating internally how to confront AI use while enabling members to benefit from it.A seed-recording app A funding program proposed an app to record the seeds that indigenous women keep. The organization’s concern was that the end users would be corporations that might develop the seeds, make them genetically modified, and sell them back to communities, so that indigenous knowledge and cultural practice drop out of the process.

Disinformation and erosion of institutional trust

Who are affected? Media organizations, electoral institutions, and the general public as consumers of a social-media-dominated information environment; public health advocates whose evidence is diluted by industry-backed materials.

Disinformation operations affecting the regionTwo large-scale, reportedly foreign-linked information operations illustrate how AI lowers the cost of producing disinformation at scale: one involving a large network of AI-generated social media personas producing region- and country-specific content, and another involving a high-volume network of websites publishing disinformation in multiple languages. Both carry potential downstream effects on training data for future language models.

Deepfakes and AI-generated content are eroding trust in media, survey institutions, and government. Foreign influence operations use large networks of AI-generated social media personas and websites publishing high volumes of content to spread disinformation. Participants cite AI-generated content in everyday feeds and in short serialized dramas; AI summaries and paid placement in search results; AI-assisted misinterpretation of health regulation; and the failure of AI tools to identify the industry funders of studies.

Participants flag these as high-risk issues that are directly linked to the health of democratic institutions and electoral integrity, compounded by an information ecosystem where most citizens get their news from platforms with opaque algorithms. Participants from the health sector framed the central issue as AI amplifying existing distortions of science: disinformation and industry influence long predate AI, but AI adds scale, speed, and an appearance of certainty. Platform algorithms that reward engagement also promote AI-generated content over human-made work.

Risks to data privacy, biometrics, and cross-border data governance

Who are affected? The general public, particularly where informed consent is unclear, or where data may later be sold or transferred to a different corporate owner; patients; students and staff whose data institutions collect; bereaved families.

Unresolved questions remain about who owns collected data if a company changes hands, how the Data Privacy Act applies, and how rights can be enforced across borders. A university data-protection officer described the routine work of checking purpose, legal basis, consent, retention, and excessive collection in contracts, and said that AI-specific guidance from the NPC, as they understood it, centers on transparency and accountability. The Data Privacy Act’s transmissibility of rights to heirs raises an unresolved question of who decides about a deceased person’s data and likeness. Subsequently, participants significantly noted that the Data Privacy Act is a limited tool for AI governance.

Risks to education and critical thinking

Who are affected? Students at all levels, new graduates entering legal and other professional fields, and educators facing increased grading and detection burdens.

Contested AI use in assessment: proctoring and plagiarism detectionStudents have been accused of cheating based on behavior flagged by AI-embedded online proctoring tools, with no clear institutional policy for contesting such findings. In a related case, a plagiarism-detection tool flagged original student work as unoriginal because an earlier, different version of the same document had previously been submitted through the system. University AI policy gaps At one university, an internal debate over whether AI use policy should sit with the IT department or with the academic side of the institution produced two parallel and ultimately unresolved policy efforts, alongside fears of over-regulation among faculty and unresolved questions of accountability in the event of a data breach involving AI tools used in teaching.

 The risk landscape is characterized by a rising number of disciplinary cases for unsanctioned student AI use; inconsistent AI policies in academic institutions; AI-embedded online proctoring tools flagging students for behavior, without clear institutional rules or channels to contest false cheating accusations; and an active debate on whether AI use is eroding students’ critical thinking and writing skills. Participants in the creative sector added reports of artists and writers feeling intimidated enough to run even their own scripts and lines through AI for correctness, and of parents and teachers being shamed for AI use even when they are experimenting and learning.

Another risk is the under-scrutinized role of institutions imposing AI tools using a top-down approach (for grading, proctoring, or instruction). Participants in the creative sector also called for AI literacy that teaches people to vet outputs and to set their own guardrails, noting that many people feel too disempowered to do so. Creative sector participants further observed that education is shifting toward a market-oriented framework driven by industry-led programming, including in the AI literacy now being taught to children.

Environmental impacts and extractive supply chains

Who are affected? Communities near proposed or existing data-center and industrial sites; indigenous peoples and farming and fishing communities in mining areas; and the public more broadly through competition for water, energy, and land. Those with no access to AI are also affected, as development proceeds around them, but without their participation or inclusion.

Planned data-center and semiconductor investment carries significant energy, water, and mineral-resource demands, set against AI’s potential benefits for environmental monitoring (forest health, crop yields, hazard mapping, energy efficiency). Participants from the environmental sector reported that there are no publicly available Philippine studies of AI’s environmental impacts, and that existing data centers do not disclose energy and water use. A resident living near an existing data center in Quezon City reported nights with no water or low pressure and noise pollution, and described the facility as built in the middle of a residential area.

The character of the Pax Silica project was itself debated. Although often discussed as a data-center project, it has been described by the Bases Conversion and Development Authority (BCDA) as an AI-driven industrial hub in New Clark City, Capas, Tarlac, backed by a framework with the United States and covering semiconductor processing, critical minerals, advanced manufacturing, and data infrastructure.^[4] Participants noted this shift in framing and characterized the project as having a trade dimension and as a hub where microchips would be manufactured, chips that, they warned, could be used not only in data centers but also in military technologies. They also noted that the national government has been vocal about making the Philippines a key player in the AI industry.

Participants working with mining-affected communities linked the national AI-hub strategy, including the Pax Silica project, to the extraction of critical minerals: in Brooke’s Point, Palawan, nickel mining communities reported noise near schools, and floods that had not occurred before, as well as the bypassing of FPIC, with no benefit to the communities themselves. In Zambales, black-sand mining and dredging have damaged beaches and farmland, and a new local road, that also damaged farmland, was described by farmers as serving miners and linking Zambales to Tarlac. Farmers added that pests now move from the damaged mountains into their fields, that they can no longer make use of once-rich soil, and that the coastal route offers direct access for shipping minerals abroad. Participants also reported that the AI-hub strategy relies on weakened consent: revised FPIC guidelines were described as easing permits for extractive industries, without consultation of community partners.

A dam built without proper environmental processA participant described a case in which a dam was built without the proper environmental assessment, the regulator investigated and found non-compliance, and the outcome was a minor fine and no demolition, which the participant called “impotent”. Participants cited this and a resort built near a protected landmark (without an assessment) as examples of why laws are reactive rather than preventive, and why this matters for the permitting of data centers.

Broader issues discussed as potential risks in the environmental sector also include the following:

• Resource demand and data gaps. Foreign studies cited by participants indicate that AI data centers drive up energy demand and require large quantities of water. Meanwhile, evidence specific to the Philippines is lacking, as information is held privately. Data gaps also extend to emissions monitoring, which relies on submissions from local governments that find such reporting difficult and give it low priority.

Participants noted that the Philippines is in the first wave of building AI data centers, which are not yet hyperscale facilities as in the United States, and cited moves in places such as New York to halt further construction; Philippine studies grounded in the context of host communities were therefore considered necessary. Participants noted that data centers can push countries to retain or increase fossil fuel usage or return to nuclear power, and that US restrictions on AI companies may lead them to relocate data centers in other developing countries, including those in Southeast Asia.

• Land, displacement, and sovereignty. Foreign reports and lawsuits describe land grabbing and displacement around data centers. Participants asked who the extraction is for, arguing that raw materials and sovereignty are being “given away” while the promised advanced technology fails to materialize. Some also expressed concern over the use of the chips and minerals in military and surveillance technologies.
• Contested benefits. Participants also debated the argument that, should the Philippines withdraw from Pax Silica, neighboring countries such as Vietnam or Indonesia would take its place. Some answered that the question is who such projects are for, and that the country need not keep pace on these terms; the environmental focus group also rated the prioritization of foreign-led development initiatives as high in both likelihood and severity of impact. A different view was held that the establishment of data centers is an opportunity to industrialize and add value if it is transparent, publicly audited, and accountable.
• Weak permitting and enforcement. Participants raised other risks about the establishment of data centers. Among them, data centers may be treated as ordinary buildings permitted by local governments with little capacity to assess environmental impact. Members of this sector also mentioned how environmental impact assessments may be subject to corruption, i.e., commissioned and paid for by the applicant. Other general enforcement issues were also raised, concerning limited oversight on data centers and reliance on reports from the public for potential violations.

• Policy frameworks for mining.^[5] They considered the Mining Act insufficient, with problems in both the law itself and its implementation, and pointed to the proposed Alternative Minerals Management Bill as a pro-community alternative. Participants also expected that the AI governance framework soon to be released by the Department of Economy, Planning, and Development (DEPDev) would not address the issues they raised.

• Indirect harms. Impacts on marine ecosystems, forestry, and biodiversity were described as indirect, arising from what AI requires (minerals, energy, cables) rather than from AI itself, which is a reason they are easily overlooked.

• Over-promotion and greenwashing. Participants warned of a risk that the benefits of development projects are over-emphasized and environmental impacts understated, a pattern reported abroad but not yet documented in the Philippines. Participants referred to a Greenpeace report on this pattern abroad and saw a need for Philippine documentation; the environmental focus group nonetheless rated greenwashing as high in both likelihood and severity.

In its own risk mapping, the environmental focus group identified three priority risks: the displacement of indigenous peoples, which also erodes their indigenous knowledge systems and practices (IKSP); the depletion of natural resources and ecosystems; and the high energy demand of AI infrastructure, which may prolong reliance on non-renewable sources. The group also rated food insecurity, adverse health impacts on affected communities, the fast decline of biodiversity and ecosystems, and the loss of land-based livelihoods as high in both likelihood and severity, while the depletion of water supply, land and soil degradation, and the lack of industrial policies were rated medium. As solutions, the group called for genuine consultation with the sectors and communities affected by data centers and AI hubs, and for government agencies to refrain from siding with large technology and mining companies.

Risks to health and public health outcomes

Who are affected? Health workers and researchers whose jobs are at risk because of automation; women and other patient groups underrepresented in the data used to train clinical AI tools; people with autism and other communities missing from training data; people outside digital systems (for example, those without social media or birth certificates) who are invisible to data-driven public health; users of health apps; and communities and countries in the Global South where public data may be used to develop harmful products that are previously not available or easily accessible to them.

An autism-awareness campaign with outdated AI suggestions A health communicator who used an AI tool to brainstorm an autism-awareness campaign received suggestions built around puzzles. When the ideas were shown to the client, they were rejected as outdated and not how the community wishes to be represented. The facilitator concluded that the communities concerned had probably never been consulted when the tool’s training data were assembled.AI and a flavor-descriptor ban A public-health advocate working on tobacco control described how counterparts relying on general-purpose AI received interpretations that a flavor ban does not extend to flavor descriptors, contrary to a Department of Justice opinion and the latest legal interpretation. The advocate had to correct what the tool had told people before making the case. The same participant noted that AI tools do not flag studies funded by the tobacco or vape industry that omit funder disclosures.

As mentioned above (“labor and economic risks”), participants from the health sector also raised risks to their labor and employment. In this sector, a health communications practitioner described how a company restructured into project-based roles without government benefits, as research teams pushed automation.

Another significant risk raised in this sector is how AI-powered clinical decision-support systems show signs of bias. This manifests on how some diagnostic markers appear to be calibrated primarily on male patient data. There are risks of misdiagnosis from biased health AI, and misrepresentations of marginalized health populations in communication.

Risks on privacy and consent also exist. Health apps and hospital or clinic agreements raise concerns over excessive collection, consent, retention, and cross-border transfers; AI-specific rules in this regard are still thin.

Health communicators also relate experiences of AI tools suggesting outdated or Western-framed approaches for communities, such as people with autism. At the same level of risk, public health advocates found AI tools giving out-of-date interpretations of existing regulations and failing to identify industry-funded studies (e.g., for tobacco control).

On the other side, participants share how some local governments use AI in tuberculosis and dengue analytics and in CCTV-based detection of violators of anti-smoking laws; how hospitals are adopting AI for efficiency gains; and how useful AI is in conducting sentiment analysis of social media to understand women’s views on family planning. AI-based monitoring (as with CCTV monitoring to enforce smoke-free rules) raises validation and misreporting concerns. Participants also recognized that AI can support the inclusion of persons with disabilities, although many still lack access to it, especially in communities without reliable internet connection.

Private capture of public health data is also a concern. A nutritionist noted that AI systems are built on public data, but pharmaceutical firms contracting developers can use them to create products that are not available to all, widening health inequity. Participants argued that data obtained free or cheaply should not simply be used for profit-seeking purposes, and that guardrails on use and pricing are needed; one participant suggested that lifestyle data could be used to create demand for products. A related risk is how the use of this data may render market-driven harm initially invisible, i.e., lifestyle and marketing data can shape health outcomes, in ways that are not recorded as health impacts.

Key feedback focused on the persistent extent of industry influence on information and policy. AI comes in the midst of a marketplace of ideas that is not level: those with resources and control of the narrative gain advantage, and AI makes it cheaper to exploit those with limited resources and power. Participants concluded, as a cross-cutting issue, that AI amplifies existing inequities and injustices, rather than creating wholly new ones.

Intellectual property, culture, and the creative sector

Who are affected? Writers, artists, illustrators, designers, filmmakers and film crew, performers, regional and community artists, and creators whose work may be used as training data; audiences and the public, who are affected by homogenized content; and indigenous communities whose designs, songs, seeds, and stories are being digitized.

Intellectual property and creative sectors (writers, artists, filmmakers, designers, performers) raised concern about AI models being trained on creative work without consent or compensation, and about platforms converting such work into commercial applications. Artists reported that once their work is in a training pool, it cannot be removed; that digital artists post their work knowing it will be scraped; and that some have returned to traditional media as a result of these concerns. Platform algorithms and the promotion of AI-generated content by large platforms have made it harder for human artists to be found. Participants state that creative work is devalued by decision-makers who do not understand what artists contribute (not only making images, but judging effectiveness and legibility). There is a concern that outputs now tend toward the average, and artists report demoralization and alienation from their own work. Participants extended this concern to artists’ mental health. As one put it, AI-generated work “just satisfies your hunger but it’s not nourishing,” and artists now ask how they can continue to develop their artistry.

Meanwhile, it was noted that film production is already deeply integrated with AI across pre-production, shooting, and post-production, with a participant stating that some AI-generated work is nearly undetectable. Participants described a debate on whether prompting is art; one view was that ideation and execution represent the ‘real’ art, and that AI outputs are not art, while another held that structuring one’s thought is a form of expression and that AI may open creative avenues for people for whom art and code were inaccessible. Among film practitioners, one view held that the industry is used to disruption, although not at this exponential pace, and that AI is a threat mainly to those who do not invest in learning it; understanding the systems, in this view, allows practitioners to retain the human intent behind the work, and the industry remains conscious of the need to preserve art as culture. Other participants stressed that AI is political even in art.

A freelance illustrator loses long-standing clients A freelance illustrator reported that income fell sharply in a year marked by a funding crisis for non-profits and by AI. An international activist alliance that had been a client since 2019 began supplying AI-generated images as references, then posted AI-generated content and commissioned much less work; followers reacted negatively. A news organization moved from per-illustration fees to a lower monthly contract, and later released an AI-generated animated video that the illustrator found lacking context for a very serious subject. The participant noted that grants and workshops for AI integration in newsrooms are plentiful.  Community arts on an island A performer and writer from an island in Mindoro described how paid commissions for local tourism and school theme songs, scripts, and graphics, which were once routine, had stopped as officials used AI instead. Because commissions are tied to local politics, challenging this would reduce the chance of being hired. The participant also reported feeling intimidated into having scripts checked by AI and said the “beauty of imperfection” is lost, along with local speech and slang.  A posthumous chatbot likeness Participants discussed a reported case in which a deceased writer’s husband used her works to make a chatbot likeness, with her voice and image, and her parents contested it. The facilitator explained that the Data Privacy Act makes data-subject rights transmissible to heirs, unlike in most jurisdictions, but does not say which heir decides, and noted that digital afterlife is already an industry elsewhere.

The use of LLMs specifically raised issues on appropriation without consent or compensation, as well as cultural homogenization and dependency. Models benefit from creators’ work while creators do not; the effect is most felt by artists with recognizable styles. Dominant models ‘flatten’ local voice, slang, and aesthetics, and make the Philippines dependent on foreign models trained on Western taste.

Participants also raised risks to cultural heritage and indigenous knowledge. While digitization of designs, languages, seeds, and sacred places can help preserve culture, it also exposes it to commercialization and corporate capture. Participants share how community elders have resisted photographing sacred places, for example, citing a history of land loss as one of the many reasons for this resistance.

Participants from regions with limited access to arts education noted that AI can give people a medium they never had, and that gatekeeping and elitism in the art world also shapes how AI is received. Policy therefore needs to protect creators without presuming that all AI-assisted creation is harmful.

Other issues

Other risks in the use of AI systems were also documented in the workshops; however, these did not represent firsthand experience or engagement by participants. These risks are nevertheless included here, and in the consolidated risk matrix, for additional context and as topics for further discussion.

Frontier AI, dual-use, and advanced-AI safety

Who are affected? The general public and national security interests, both from potential misuse by malicious actors and from the more speculative but flagged risk of highly capable future AI systems.

Adversarial prompting of LLMs can produce harmful instructions, and AI systems have been documented exhibiting deceptive or goal-directed behavior beyond simple hallucination, according to some participants. Participants raised concerns that semiconductors and minerals extracted in the Philippines could be used in military and surveillance technologies. Participants also raised the question of liability and accountability in automated operations, such as satellite operations, with some raising concerns on how AI-enabled weapons systems have been advertised and marketed abroad.

Consolidated risk matrix

All ratings were prescribed by respective groups from each sector during the workshop exercises.

Risk area	Affected groups	Likelihood	Severity	Illustrative example
Labor intensification (more workload for the same or less pay) and job displacement in BPO/IT	Gig workers, BPO/IT employees, SMEs	High	High	Clark QA layoffs; ~1,500 BPO employees affected; Content moderation restructuring
Displacement, devaluation, and intensification of creative and communications work	Freelance illustrators and designers, film crew, marketing staff, community artists, health communicators	High	High	Camera-assistant and editing roles disrupted by AI tools; freelance clients shifting to AI images; marketing department outsourced as AI introduced; provincial art commissions lost; income loss, entry-level displacement of professionals, loss of job security, and the devaluation and demoralization of artists
Displacement of entry-level health jobs and health expertise	Entry-level health workers, health professionals and experts	High	High	Nutritionists and researchers as examples of jobs threatened
Lack of transparency in AI model training and bias	General public, LLM users, customer-service recipients	High	High	Major technological platforms cited as examples of opaque training and unknown bias
Unreliable, outdated, or culturally mismatched AI outputs in health and policy work	Health communicators, public-health advocates, communities missing from training data	High	Medium	Autism campaign ideas judged outdated by the community; AI misreading of a flavor-descriptor ban; industry-funded studies not flagged
Amplification of commercial-industry tactics affecting health	Public-health advocates, consumers, health regulators	High	Medium	AI magnifies food and tobacco industry disinformation and makes science confusing; AI-powered tools used for other purposes (e.g., restaurant systems) that affect health outcomes
AI-generated child sexual abuse material and image-based abuse	Children, girls, LGBTQIA+ youth, women	High	High	A 26,385% rise in AI-generated child sexual abuse videos was reported by the Internet Watch Foundation in 2025; “nudify”-app use is documented; deepfake pornographic material on porn sites and secure messaging apps
AI-enabled technology-facilitated gender-based violence (TFGBV) and its structural drivers	Women, girls, trans women, LGBTQIA+ people	High	High	AI-enabled TFGBV (e.g., gendered disinformation), with causes identified as AI frameworks and policies that are not gender-responsive, low AI awareness and literacy, underrepresentation of women in AI, lack of platform accountability, and inherent bias and prejudice in AI models
Technology-enabled grooming and early marriage of indigenous girls	Indigenous girls and young women, mothers and families	Medium	High	Girls of about 15 meeting adult men through proximity-based apps and (Facebook) Messenger; later abandonment and abuse
Discrimination and policing of trans and gender-diverse people	Trans women and gender-diverse people	High	High	Deepfake “trans-baiting” videos; AI-based checklist for identifying trans women; identity-document mismatch at banks and borders; facial-recognition concerns
Platform enforcement that suppresses advocacy	Women’s rights and LGBTQIA+ advocates, safe-abortion advocates	High	Medium	Automated takedowns triggered by flagged words; Western-framed community guidelines
AI-enabled scams targeting women and LGBTQIA+ people	Women, LGBTQIA+ people	High	Medium	Scams identified by the women and gender focus group; low AI awareness and literacy among targeted communities
Disconnect between AI use and advocacy	Women’s rights and gender advocates, CSOs	High	Low-Medium	Use of AI by NGOs
Digitization of culture and creativity (preservation vs. resistance)	Indigenous communities and indigenous women, cultural and community organizations	High	Medium	AI-generated rally and welcome songs; proposed seed-recording app; elders resisting the photographing of sacred places
Surveillance and transnational repression of activists	Human rights defenders, journalists, activists	High	High	AI-amplified red-tagging; digital harassment; open-source data used to profile and track
Government AI surveillance for public-health enforcement	General public, communities under surveillance	Medium	Medium	Video surveillance for smoking violations; health FGD participants noted that government use of AI is less regulated
Erosion of institutional trust through disinformation	Media, electoral institutions, government	High	High	Foreign influence operations (including AI-generated social media persona networks and high-volume disinformation website networks) compound deepfake-driven distrust; AI summaries in search results
Cognitive disempowerment / erosion of critical thinking	Students, new graduates, educators, children and young people	High	Medium-High	Rising disciplinary cases for unsanctioned AI use; an exam-performance study among senior high students; emotional coping using AI
Data privacy and cross-border data governance gaps	General public, biometric-data subjects, patients	Medium	High	DNA/genetic profiling services; biometric iris-scan enrollment; unclear data ownership on company sale; health-app consent; posthumous chatbot likeness
Private capture of public data and widening health inequity	Patients, Global South communities	High	High	Public and free data used by pharmaceutical firms through AI developers to create products not available to all; hospital genetic data used for drug development
Environmental impact of AI/data-center buildout	Communities near data-center sites; general public, food-insecure and resource-dependent communities	High	High	Energy, water, and mineral demand of proposed data centers; outdated Environmental Impact Assessment framework; reported water outages and noise near an existing data center
Extractive supply chain, displacement, and weakened consent tied to the AI-hub strategy	Indigenous and mining-affected communities, women on the front lines, environmental human rights defenders (EHRDs)	High	High	Nickel mining in Palawan and black-sand mining in Zambales; revised FPIC guidelines criticized; Executive Order No. 122 rejected by affected communities; inconsistent jobs and affected-family figures for Pax Silica
Greenwashing and prioritization of foreign-led development initiatives	Host communities, indigenous peoples, general public	High	High	Benefits of AI-hub and data-center projects over-emphasized and environmental impacts understated; argument that neighboring countries will take Pax Silica if the Philippines withdraws
Depletion of water supply and land (soil) degradation	Farming communities; communities near mining and data-center sites	Medium	Medium	Damaged farmland and soil in Zambales; water outages near an existing data center in Quezon City
Weak technology transfer and industrial policy in AI investment	Local workers, industries, and host communities	Medium-High	Low-Medium	No genuine capacity building from “tech transfer” in the Philippine context; lack of industrial policies
Dual-use / uplift risk for malicious actors	General public, national security	Low-Medium	High	Adversarial prompting of LLMs for harmful instructions; AI-enabled scam operations in the region
Bias in health-related AI systems	Women and other underrepresented patient groups	Medium	Medium-High	Clinical decision-support tools trained on data skewed toward male patients, raising misdiagnosis concerns; exclusion of marginalized communities from health data
Weak accountability for cybersecurity and election-related AI use	Voters, electoral candidates, general public	Medium	High	COMELEC rules on AI in campaign material; comparable cases from Indonesia, Nepal, and Bangladesh elections
Intellectual-property, cultural, and compensation gaps for creators	Writers, artists, filmmakers, journalists, indigenous knowledge-holders	High	High	Creative work used to train models without consent or compensation; scraping of posted work; digitization of indigenous designs and seeds; algorithmic promotion of AI content; voice and likeness cloning
Platform capture, AI slop, and homogenization of creative markets	Artists, actors, creators, audiences	High	High	Tobacco ads on AI micro-dramas, homogenization (“sameness”) of outputs, and erosion of cultural transmission
Autonomous AI generation without human intervention	Creative workers, audiences	Low-Medium	High	N/A
Bargaining-power shift away from creative workers	Freelance and contract creative workers	Low	Low	Clients as top beneficiaries; per-illustration fees replaced by lower monthly contracts

 IV. Recommendations

The recommendations below come from participants in both workshops and from the risks set out in the sectoral risk map and the consolidated risk matrix. They were voiced in sectoral discussions and focus group discussions, and have been consolidated, grouped, and lightly edited for clarity. They reflect the views raised by participants and should not be read as a consensus position of every organization that took part.

Many other recommendations do not need a new law to be implemented: executive agencies, regulators, companies, and communities can act now under existing mandates, including the Data Privacy Act, the environmental impact assessment system, the Labor Code, and laws on violence against women and children.

Several threads run through the recommendations.

First, transparency is a precondition for everything else: civil society, regulators, and affected communities cannot assess or contest AI harms they cannot see, whether these concern training data, data collection, energy and water use, or AI-assisted decisions.

Second, accountability should sit where power and knowledge sit, i.e., with upstream model developers, platforms, and large deployers, rather than falling only on local companies, individual workers, or end users.

Third, the people most exposed to AI-related harm (women and girls, LGBTQIA+ people, indigenous peoples, workers, artists, patients, and human rights and environmental defenders) should have a guaranteed place in decisions about AI.

Fourth, oversight needs institutions with a clear legal mandate and permanence; participants warned that task forces which are never institutionalized have not worked.

Finally, priority should go to the risks that participants rated high in both likelihood and severity, including technology-facilitated gender-based violence, surveillance of activists, labor displacement, disinformation, the private capture of public data, and the extractive supply chains tied to the AI-hub strategy.

The recommendations are organized by the actor best placed to act upon them: (A) Congress, (B) executive agencies and regulators, (C) companies, platforms, and AI developers, (D) civil society, communities, educational institutions, and professional bodies, and (E) responsibilities shared by all of these actors.

A. Recommendations to Congress

Congress, and in particular the Technical Working Group consolidating the AI bills, is in a position to set the basic architecture of AI governance in the Philippines: who oversees AI, which uses are treated as high risk, who is liable when AI causes harm, and how existing laws on labor, violence, privacy, and intellectual property apply to AI. The recommendations in this part are largely about building these protections into the substitute bill and into amendments to existing laws.

1. Give the AI oversight body created by the substitute bill a clear legal mandate, a clear lead agency, and permanence. (Cross-sector)

Participants described a fragmented and reactive regulatory landscape, in which it is unclear whether the DICT, DOLE, the NPC, or another agency leads on AI. The bills now before the Technical Working Group propose different bodies (an AI board, bureau, authority, or commission) under different departments. Whichever model is chosen, the law should define the body’s mandate over standards, complaints, and coordination across agencies, and give it a permanent institutional footing. Environmental sector participants cautioned that bodies that are not institutionalized have not worked in their experience.

2. Expand the high-risk categories in the bill to cover AI that affects vulnerable groups, AI used in grading, assessment, and the rejection of applications, and AI in healthcare. (Cross-sector; labor, education, health)

The high-risk categorization in the bills does not yet reflect the harms documented by participants. The list should explicitly cover AI systems that affect children, neurodivergent people, and other vulnerable groups; systems that grade or assess students and workers or reject applications for jobs, loans, or services; and AI used in healthcare, including government uses for monitoring and analytics. These categories should carry correspondingly stricter requirements, such as risk assessments tailored to context before deployment, and stronger penalties.

3. Place accountability upstream on the providers of large language models and other foundation models, and consider defining AI as a product under Philippine law. (Cross-sector; labor)

At present, liability tends to fall on the local company deploying an AI tool, even though systemic risks in a model can only be fixed by its developer. Patterned on approaches in other jurisdictions, the law should assign obligations to upstream model providers by requiring a local representative to be created and registered before such provider may operate in the country. Treating AI as a product under the Consumer Act (Republic Act No. 7394) would also allow people harmed by it to bring product-liability claims.

4. Extend AI obligations proportionately to small and medium enterprises rather than exempting them altogether. (Labor; SMEs)

SMEs are both vulnerable to market pressure driven by AI and potential sources of harm if their use of AI is unregulated. Rather than exempting them from the requirements that apply to larger corporations, the law should apply scaled obligations, such as privacy disclosures and employee protections, that SMEs can realistically meet.

5. Require the identification of high-compute AI use and impose a compute surcharge that finances a welfare fund for workers outside existing protections.(Labor)

Gig workers and SME employees are often not covered by the protections available to employees of large BPO companies and corporations. A surcharge on high-compute AI use (e.g., measured through a revenue or usage threshold), channelled into a dedicated welfare fund, would make those who profit most from AI contribute to protecting the workers most exposed to displacement.

6. Modernize labor and social protection for workers displaced by AI. (Labor; creative sector, health)

The Labor Code and standard BPO contracts should be reviewed in light of the layoffs and restructuring documented by participants, including the BPO layoffs and content-moderation cases. Separation and unemployment compensation should be expanded beyond the currently proposed extension (in the House Technical Working Group’s consolidated draft of the AI Development and Regulation Act) of the notice period from 30 to 60 days, and SSS coverage for workers displaced by AI should be clarified. These protections should also reach freelancers and below-the-line crew in the creative sector, and entry-level health workers, whose displacement participants rated as high in both likelihood and severity.

7. Amend the Safe Spaces Act, the anti-OSAEC law, and the Anti-Violence Against Women and Their Children Act (RA 9262) to cover AI-generated and technology-facilitated violence, and clarify which law applies to which case. (Gender; children; LGBTQIA+)

Deepfakes, AI-generated sexual abuse material, and other forms of AI-enabled technology-facilitated gender-based violence are outpacing the law. Participants rated these risks among the most likely and most severe of all. Survivors also struggle to file cases because the existing laws overlap and it is unclear which one applies. Amendments should cover AI-generated content explicitly and give survivors a clear legal pathway. Women’s groups also reported that they have largely not been consulted on AI bills that are not specifically about gender; the Technical Working Group should seek their input on the substitute bill as a whole, not only on its gender provisions.

8. Enact legal gender recognition with privacy safeguards, and ensure that AI-assisted document verification does not penalize trans people. (LGBTQIA+)

Trans and gender-diverse people face mismatches between their identity documents and their appearance at banks and borders, a problem that automated and AI-assisted verification can make worse. Legal gender recognition should include privacy safeguards, such as no public notice of a change of name or gender marker, so that the process does not itself expose people to outing.

9. Clarify copyright, consent, and compensation for the use of creative work as AI training data. (Creative sector; intellectual property)

Artists reported that work used in a training pool cannot be removed and that creators do not benefit from it. Intellectual property law should make clear when consent is required for the use of creative work in training, how creators can withhold or withdraw it, and how they are to be compensated. Workshop participants rated training-data appropriation and voice and likeness cloning as high-severity risks.

10. Clarify under the Data Privacy Act who may exercise data-subject rights over a deceased person’s data and likeness. (Data privacy; creative sector)

The Act makes data-subject rights transmissible to heirs, but sets no hierarchy among them. This gap became visible in a case in which a deceased writer’s works were used to create a chatbot likeness. The law should set out who may consent to, or object to, the use of a deceased person’s data, voice, or likeness.

11. Conduct a congressional inquiry into AI-related harms and regulatory gaps, and hold budget hearings on funds earmarked for worker upskilling. (Cross-sector; labor)

Data on AI-related harms are currently scattered across agencies. A congressional inquiry would bring them together and create public accountability. Budget hearings should also examine how funds allocated for worker upskilling are actually being used.

12. Review the Mining Act and Executive Order No. 122 in light of the demand for minerals created by AI infrastructure, and consider the proposed Alternative Minerals Management Bill. (Environment; indigenous peoples)

The supply chain of AI begins with the mining of critical minerals. Mining-affected communities found the Mining Act insufficient in both its text and its implementation, and rejected Executive Order No. 122, which treats critical-minerals projects as national priorities, for presupposing a “responsible” mining that communities have not experienced. Congress should assess whether existing mining law protects host communities and indigenous peoples from the additional pressure that AI infrastructure creates, and consider the Alternative Minerals Management Bill, which participants supported as a pro-community measure.

B. Recommendations to executive agencies and regulators

Executive agencies and regulators do not need to wait for a new AI law to act on the risks raised by stakeholders. Much of what workshop participants cited as recommendations fall within existing mandates: the NPC’s authority over the processing of personal data, the environmental impact assessment system, the education agencies’ authority over curricula and assessment, and the planning of the national AI roadmap. This part sets out what these bodies can do now, and how they can prepare for the institutions a new law would create.

1. Form a multi-agency task force on AI now, with a mandate to map the actors and risks across the AI pipeline and a defined path to becoming a permanent body. Cross-sector)

Participants noted that currently, national AI planning follows an evolving, multi-agency approach, including successive versions of a national AI roadmap associated with the Department of Trade and Industry (a National AI Strategy Roadmap 2.0 — Philippines[6]). There are also discussions on whether the Department of Economy, Planning, and Development (DEPDev, formerly the National Economic and Development Authority, NEDA) should take a coordinating role. Participants reported that DEPDev was preparing to release an AI governance framework, but they expected it to be general and unlikely to address the concerns raised in the workshops.

Which agency currently owns the roadmap, and what its components are, remains unclear. Participants also found the mandates of the environment and ICT agencies unclear in relation to AI, and the Department of Information and Communications Technology (DICT) was seen by some as facilitating investors rather than regulating.

Thus, pending legislation, a task force should create institutional capacity where none exists. Its first task is to identify the key actors across the AI pipeline (developers, deployers, and the populations at risk) and the specific risks each poses. The Climate Change Commission, which began as a task force, offers a precedent. To avoid the fate of task forces that were never institutionalized, its mandate and its path to permanence should be defined from the start.

2. Establish a multi-sectoral oversight council with guaranteed seats for the groups most exposed to AI-related harm. (Cross-sector)

Decisions on AI policy and implementation should not be dominated by corporate interests. Seats should be guaranteed for women, youth, academe, media, and labor, as well as for indigenous peoples, LGBTQIA+ communities, and the health, environmental, and creative sectors.

3. Require AI companies to disclose what data they collect, how they acquire it, and how they use it. (Cross-sector; data privacy)

Participants repeatedly described AI systems as a “black box”: the lack of transparency in model training and data practices was rated high in both likelihood and severity. Mandatory disclosure would close the information gap that civil society, regulators, and the public now face, and is a precondition for most other forms of accountability.

4. Strengthen the NPC’s AI rules so that they protect people, not only restrict and penalize. (Data privacy; health; gender)

Participants found the Commission’s current circular on AI vague and shallow on protection. The rules should add concrete protection mechanisms for citizens, and AI-specific guidance on consent, data retention, and cross-border transfer. Health apps, clinics, and hospitals should be required to disclose in their contracts and terms what data they collect. Participants also flagged over-collection, retention, the accountability of large technology companies, and the question of who gets access to publicly funded health data.

5. Build gender analysis and protection mechanisms into the national AI roadmap and the forthcoming AI governance framework. (Gender; LGBTQIA+; indigenous women)

The national AI roadmap was reported to contain no gender analysis beyond a general statement of benefit. It should address harms to women, LGBTQIA+ people, and indigenous women specifically, including the structural causes identified by the women and gender focus group: frameworks that are not gender-responsive, low AI literacy, the underrepresentation of women in AI, weak platform accountability, and bias in AI models. Consultations should include ordinary survivors, not only high-profile cases. The AI governance framework expected from the DepDEV should likewise be reviewed for these gaps, with women’s groups consulted, before its release.

6. Launch a whole-of-society AI literacy campaign through an executive order, with an interagency body and the DICT as a possible lead agency. (Education; cross-sector)

The campaign should reach students, teachers, parents, and adult learners, not only the education sector, and avoid a framing focused only on children. Participants stressed that it should also reach people who do not know they are already using AI. Low AI awareness and literacy was identified as one of the root causes of technology-facilitated gender-based violence.

7. Regulate high-risk uses of AI in education and strengthen foundational skills in the curriculum. (Education)

Automated grading, assessment, and proctoring should require justification and, absent clearer standards, be discouraged. The curriculum should include a course or module on AI ethics, potentially beginning at the basic education level, and AI-free components so that students develop core competencies before relying on AI tools.

8. Assess AI in health before it is deployed, provide validation and redress for AI-based public-health monitoring, and guard against industry influence over health AI policy. (Health)

Risk assessments should be tailored to context and applied before deployment, including to government uses such as monitoring, analytics, and surveillance for public-health enforcement. AI-based monitoring by local governments should have mechanisms to validate and correct misreports, and people outside digital systems should not be excluded from data-driven public health. Guardrails are also needed so that large technology companies and the global food, tobacco, and pharmaceutical industries do not shape the AI policies that apply to them; participants warned that AI can magnify industry disinformation and make science harder to follow.

9. Require benefit-sharing or cost-reflective pricing when public, free, or patient data are used to develop commercial products. (Health; data governance)

Participants described public and hospital data, including genetic data, being used by pharmaceutical firms and AI developers to create products that are not available to all. Workshop participants rated the use of public health data for profit-oriented products as high in both likelihood and severity. Benefit-sharing would help ensure that the public, whose data make these products possible, also benefits from them.

10. Require data centers and other AI infrastructure to disclose their energy, water, and land use, and commission studies of their effects on surrounding communities. (Environment)

No Philippine data exist on the environmental footprint of AI, and operators do not disclose it. Participants reported water outages and noise near an existing data center, and rated the energy and water demand of AI infrastructure, the decline of biodiversity, food insecurity, and adverse health impacts as high in both likelihood and severity. Disclosure and independent studies are needed before further buildout.

11. Clarify which agency regulates AI infrastructure, and build the capacity of local governments and regional environmental offices to review data-center permits. (Environment)

Participants found the mandates of the environment, energy, and ICT agencies over AI infrastructure unclear, and local permitting capacity limited; data centers are currently permitted much like ordinary buildings. Agencies should clarify their mandates and consider whether an AI-specific administrative order is needed. Processes between local governments and national agencies should also be consistent and streamlined, with interoperable systems for review and monitoring.

12. Ensure that environmental impact assessments are independently conducted or verified, that enforcement is preventive, and that the public can report violations. (Environment)

Studies commissioned and paid for by applicants tend to favor them, and participants described enforcement as reactive and weak. The environmental impact assessment framework, which participants considered outdated for AI infrastructure, should be updated. Because the environment department acknowledges that its oversight is limited and relies on public reports, it should provide channels for reporting violations, including anonymously, and respond to them.

13. Consult affected communities before approving AI-related projects, uphold free, prior and informed consent, and publish consistent figures on jobs and affected families. (Environment; indigenous peoples)

Participants criticized the revised FPIC guidelines and pointed to inconsistent figures on jobs and affected families for projects such as Pax Silica. Mining for the minerals that AI infrastructure depends on, such as nickel mining in Palawan and black-sand mining in Zambales, has been linked to displacement, the loss of land-based livelihoods and indigenous knowledge, and the criminalization and surveillance of communities and environmental defenders, all rated high in likelihood and severity. The economic, social, and cultural benefits of such projects should be set out in advance and be verifiable. Participants also called on national agencies and local governments not to side with large technology and mining companies: permits should be granted only after companies have shown due diligence in meeting all requirements, and where affected communities resist, local governments should stand with them.

14. Make technology transfer and the training of Filipino engineers a condition of agreements with foreign investors, and subject AI-hub projects to transparency and public audit. (Environment; industrial policy)

Participants argued that the Philippines should benefit from the AI industry, not only host it. They observed that “tech transfer” has so far brought little genuine capacity building, and that the country lacks industrial policies to capture the benefits of foreign-led development initiatives. Participants also questioned whether the country needs to compete with its neighbors for such projects at all, and asked that their benefits to Filipinos be demonstrated first.

15. Revisit the regulation of private drones, which are currently treated as toys even where they are used to intimidate. (Environment; human rights defenders)

Participants working with mining-affected communities reported drones being used as tools of aggression or intimidation, part of a broader pattern of surveillance of communities and environmental human rights defenders. They called for regulation of the use of such drones.

16. Support creative workers through disclosure of AI use in commissioned work, upskilling, social protection, and funding for community arts outside Manila. (Creative sector)

Clients and agencies, including government agencies procuring creative work, should disclose when they use AI, building on the practice reported among some advertising agencies. Participants described unpaid retraining costs, shrinking commissions, AI replacing local commissions, and the lack of institutions that train below-the-line crew. Public support for upskilling, social protection, and community arts spaces, especially in the provinces, would respond to the income loss and demoralization that workshop participants rated high in both likelihood and severity.

17. Explore Philippine-specific small language models and corpora, while protecting the data of children and families. (Creative sector; culture; cross-sector)

Dependence on foreign frontier models trained largely on Western material risks homogenizing outputs and eroding Filipino identity and local voice. Locally built models could reduce this dependency, provided they are developed with strong data protection and with the consent of the communities whose language and culture they draw on.

C. Recommendations to companies, platforms, and AI developers 

Companies that develop, deploy, and host AI hold most of the information and much of the power that determine how AI affects people. Participants observed that developer behavior is largely reactive to regulation and driven by profit, and that self-regulation by foreign companies is not enough. The recommendations in this part set out what companies should do regardless of what the law eventually requires.

1. Declare AI use across its lifecycle, provide disclosures to deploying companies and end users, and keep transparency logs of AI-assisted decisions. (Cross-sector; labor)

Lack of transparency was identified as a root cause of several risks in the matrix. Companies should formally declare where and how they use AI, monitor its downstream consequences, and keep logs of AI-assisted decisions. Developers should give deploying companies and end users the information they need to understand the limitations and risks of their systems.

2. Form a local AI developers’ association, adopt a code of conduct, and submit to independent third-party audits. (Cross-sector; creative sector)

An association and code of conduct would build accountability infrastructure specific to the Philippine market rather than relying only on the self-regulation of foreign companies. Audits should be independent rather than self-conducted, and should pay particular attention to the use of intellectual property and creative work as data.

3. Be transparent about training data, and give creators ways to opt out and to be compensated. (Creative sector; intellectual property)

Platforms and developers benefit from creative work without paying the people who made it. Workshop participants rated training-data appropriation, voice and likeness cloning, and platform algorithms that favor AI-generated content over human work among the most severe risks to the creative sector.

4. Adopt internal safety policies and commit to engaging with civil society watchdogs and independent safety experts. (Cross-sector; education)

Given developers’ profit motive and the risk of regulatory capture by large technology companies, external scrutiny is essential. Companies should commit in advance to listening and responding to watchdogs and independent experts, rather than acting only when regulation forces them to.

5. Apply community guidelines consistently and with local context, and review automated takedowns that suppress advocacy. (Gender; LGBTQIA+; civic space)

Women’s rights, safe-abortion, and LGBTQIA+ advocates reported takedowns triggered by flagged words and community guidelines framed around Western contexts. Platforms should improve how they implement their existing guidelines, review automated enforcement that removes legitimate advocacy, and act more effectively against AI-enabled abuse, gendered disinformation, and scams that target women and LGBTQIA+ people.

6. Train developers on sexual orientation, gender identity, and diversity, and move beyond binary data that erase gender-diverse (LGBTQIA+; gender)

AI inherits the biases of its developers and its data. Systems built on binary, sex-disaggregated data erase trans and gender-diverse people and can enable their discrimination and policing, as in the AI-based checklist for identifying trans women reported by participants.

7. Test health communication tools with the communities concerned, and disclose their training data and limitations. (Health)

An autism awareness campaign whose AI-generated ideas the community judged as outdated showed what happens when communities are missing from training data. Testing with affected communities and honest disclosure of limitations would reduce the risk of unreliable or culturally mismatched outputs in health and policy work.

D. Recommendations to civil society, communities, educational institutions, and professional bodies

Civil society organizations and communities are often the first to see AI-related harms, and many of the protections that participants called for depend on their own organizing. This part covers what communities, schools, professional associations, and civil society can do, often with support from government and donors, to build literacy, set standards, protect their members, and gather the evidence needed for advocacy and redress.

1. Deliver digital and AI literacy training at the community level, through schools, parent-teacher associations, and barangay-level trainers. (Education; data privacy; gender)

A “training of trainers” approach can carry AI and digital literacy to communities that national campaigns may not reach, including women and LGBTQIA+ people targeted by AI-enabled scams. Existing but under-enforced Data Privacy Act mechanisms at the barangay level, such as designated Compliance Officers for Privacy, offer a practical local entry point.

2. Teach people to vet AI outputs and to set their own guardrails, without leaving the responsibility for literacy to users alone. (Education; creative sector)

Creative sector participants observed that models are better at material that is accessible or paid for, so information asymmetry is as much a problem as misinformation. They described personal guardrails, such as keeping one’s own writing in a closed project folder and not using general chat tools for publishable work, but noted that many people feel too disempowered to set limits.

3. Develop guild and professional-association standards and collective protections for creative workers that go beyond individual pledges. (Creative sector)

A manifesto produced by a design studio was seen as a good starting point but too weak to protect artists on its own. Collective standards, for example on disclosure of AI use, would give creative workers more bargaining power than individual commitments. Such standards should also protect workers’ autonomy over their craft where employers mandate the use of AI.

4. Apply free, prior and informed consent and community benefit to any digitization of indigenous designs, languages, seeds, or sacred places, and respect decisions not to digitize. (Culture; indigenous peoples; creative sector)

Digitization can preserve culture, but it can also expose it to commercialization and capture. Participants described elders resisting the photographing of sacred places for fear of their exploitation, and the commercial reuse of indigenous designs. The women and gender focus group framed this as a tension between preservation and resistance.

5. Support community-owned, locally controlled technology so that women can share information and report violence without depending on large platforms. (Gender; LGBTQIA+)

Participants described the dilemma of needing platforms for advocacy while being targeted on them. Tools that work on local networks or offline would reduce this dependency. If AI tools for survivors are built, they should guarantee privacy, referral to a human, and an emergency pathway. Participants also asked who would develop such tools, noting that AI development remains largely male-led, and stressed the need to trace the supply chain behind any tool offered to survivors.

6. Consider collective, evidence-based strategic litigation against platforms. (Gender; indigenous peoples)

Drawing on cases brought on behalf of children and young people, participants suggested building a pool of evidence on harms to women and indigenous peoples that could support collective legal action.

7. Carry out baseline research on technology-facilitated early marriage and exploitation of indigenous girls, and on awareness of AI among rural and indigenous women. (Gender; indigenous peoples; children)

Participants described indigenous girls of about 15 meeting adult men through proximity-based apps and Messenger, often followed by abandonment and abuse. The organization working with these communities has no baseline data and requested for help in gathering it.

E. Shared responsibilities across all actors

Some recommendations cannot be assigned to a single actor. AI changes quickly, its harms cross sectoral lines, and responsibility for them is spread along the supply chain. The final recommendations concern how accountability and consultation should work across all of the actors above.

1. Hold every actor in the AI supply chain accountable for its part in AI’s impact. (Cross-sector; education)

Accountability cannot rest with a single actor. Developers, deploying companies, institutions that require AI use (such as schools and employers), and individual users each carry responsibility in proportion to their power and knowledge.

2. Institutionalize continuing consultation across the supply chain, rather than treating consultation as a one-time event. (Cross-sector; labor, education)

Because AI keeps evolving, consultation should be ongoing and should cover both upstream developers and downstream deployers. Mechanisms could include a technical working group, regular regulatory circulars, and standing committees with representatives of higher education institutions, following the model of CHED’s periodic consultation with schools.

3. Pursue sector-by-sector “small wins” alongside comprehensive legislation, with continuing consultation of practitioners and affected people. (Health; cross-sector)

Health participants favored a sectoral approach for regulation, noting that earlier attempts stalled on disagreement over issues. Because perceptions of risk are shaped by personal experience, consultation should continue to involve practitioners and the people affected, such as patients, as policy develops.

 

__________________________

[1] Proposed AI bill gains momentum in Congress, PhilStar.com, 14 July 2026, at https://www.philstar.com/business/technology/2026/07/14/2542095/proposed-ai-bill-gains-momentum-congress

[2] Estopace, E. “DOST-ICTO transitions to Department of ICT”, PhilStar.com, 13 June 2016, at

https://www.philstar.com/business/technology/2016/06/13/1592211/dost-icto-transitions-department-ict

[3] Barreiro, V. (24 March 2026) “Amount of AI-enabled child sexual abuse imagery increased in 2025 – report”, Rappler, at https://www.rappler.com/technology/internet-culture/internet-watch-foundation-ai-child-sexual-abuse-imagery-report/

[4] Bases Conversion and Development Authority, as reported by PTV News, “BCDA debunks ‘myths’ concerning Pax Silica project.” https://ptvnews.ph/bcda-debunks-myths-concerning-pax-silica-project/

[5] Executive Order No. 122 (18 August 2026), National Policy Framework for Developing the Critical Minerals Industry. See https://www.officialgazette.gov.ph/2026/08/18/executive-order-no-122-s-2026/

[6] See NAISR 2.0 – National AI Strategy Philippines at https://naisr.cair.ph/.