Partnership on AI
AI-generated text. This page was generated using artificial intelligence.
Partnership on AI (PAI) is a nonprofit partnership of academic, civil-society, industry and media organizations. It develops practical guidance through research and collaboration and seeks to inform public policy and professional practice. It describes itself as neither a trade group nor an advocacy organization, distinguishing its convening role from representation of a single industry's position.[1]
Role in policy and governance
PAI's work includes public-policy coordination, AI and human connection, labor and the economy, and safety-critical AI. Its policy program seeks connections between regulatory frameworks and practical implementation, including international coordination and participation by different sectors. Its labor program brings economists and worker representatives into discussions with partner organizations.[3]
These programs address different governance problems: the effects of AI on relationships and information, distribution of economic benefits, and safety in consequential applications. Participation in a working group does not necessarily mean that an organization adopts every PAI recommendation. Particular frameworks identify their intended users and the practices they recommend.[3]
Responsible Practices for Synthetic Media
PAI's synthetic-media framework provides voluntary guidance for technology builders, content creators, and distributors or publishers. It principally addresses audiovisual material and distinguishes direct disclosure, such as labels visible to an audience, from embedded information, such as metadata or cryptographic provenance. The recommendations recognize that synthetic content can have legitimate uses as well as cause harm.[4]
The framework calls for accessible policies, appropriate consent and responses when harmful synthetic media is distributed. It distinguishes publishers exercising editorial control from platforms hosting third-party material. PAI expressly says it does not audit or certify participating organizations under this framework. A company's participation therefore should not be described as certification, legal compliance or proof that every item of its content follows the guidance. The document is designed to evolve through review and practical case examples.[4]
Foundation-model deployment guidance
PAI's deployment guidance tailors recommendations to a model's capabilities, its release method and whether an update significantly expands its functions. It distinguishes narrow-purpose models, more general systems and frontier capabilities, as well as open access, restricted interfaces, closed development and research release. These are categories for using the guidance, not universal legal classifications.[5]
The framework recommends processes for identifying emerging risks, assessing security vulnerabilities and revisiting governance as capabilities change. It emphasizes that downstream application developers still need to assess particular uses. Its practical contribution is to organize safety work around decisions throughout development and deployment, rather than presenting one undifferentiated checklist for every model. It is intended to complement regulatory approaches and remains a living resource.[5]
Funding and governance
PAI discloses philanthropic support and corporate charitable contributions. Its governance page identifies original multiyear grants from Apple, Amazon, Meta, Google/DeepMind, IBM and Microsoft. It says primary corporate funding is general operating support rather than earmarked support exchanged for services or particular outputs.[2]
Its published bylaws description requires equal numbers of directors representing for-profit and nonprofit organizations. The page assigns directors fiduciary and strategic oversight rather than authority to determine project findings. These are PAI's stated institutional safeguards; they should not be treated as an independent finding that influence concerns cannot arise. A partner relationship, a board seat and financial support are distinct relationships and should be identified separately when discussing a specific contribution.[2]
Selected contributions
- Responsible Practices for Synthetic Media: disclosure, consent and distribution guidance.[4]
- Guidance for Safe Foundation Model Deployment: practices tailored to capability and release choices.[5]
- Policy coordination and research programs linking governance with labor, economic and social effects.[3]
Related articles
- Policy
- Public Citizen
- Frontier Model Forum
- Google DeepMind
- Stanford Institute for Human-Centered Artificial Intelligence
References
- ↑ 1.0 1.1 Partnership on AI, About Us. Accessed September 12, 2026.
- ↑ 2.0 2.1 2.2 Partnership on AI, Transparency and Governance. Accessed September 12, 2026.
- ↑ 3.0 3.1 3.2 Partnership on AI, Our Work. Accessed September 12, 2026.
- ↑ 4.0 4.1 4.2 Partnership on AI, Responsible Practices for Synthetic Media. Accessed September 12, 2026.
- ↑ 5.0 5.1 5.2 Partnership on AI, Guidance for Safe Foundation Model Deployment. Accessed September 12, 2026.