Partnership on AI (PAI) is a multi-stakeholder 501(c)(3) nonprofit, headquartered in San Francisco, California, founded in 2016 by Amazon, Apple, DeepMind/Google, Facebook (now Meta), IBM, and Microsoft to develop best practices, share research, and inform public policy on AI. It was the canonical multi-stakeholder venue for AI ethics discussion in the late-2010s; its policy salience declined after 2020 as frontier labs built in-house policy teams. It is led by CEO Rebecca Finlay, in post since 2022, and is known for its multi-stakeholder convening on AI ethics, its media and disinformation work, and its synthetic-media framework.
| Field | Value |
|---|---|
| Type | 501(c)(3) multi-stakeholder nonprofit |
| Headquarters | San Francisco, California |
| Founded | 2016 |
| Founding members | Amazon, Apple, DeepMind/Google, Facebook (now Meta), IBM, and Microsoft |
| Current CEO | Rebecca Finlay (since 2022) |
| Known for | Multi-stakeholder convening on AI ethics; media and disinformation work; synthetic media framework |
History
PAI was founded in 2016 as an industry-led consortium to develop best practices, share research, and inform public policy on AI. The six founders were joined by a growing roster of nonprofit, academic, and civil-society partners including the ACLU, Human Rights Watch, OpenAI (early), UNICEF, and others. Between 2017 and 2019 PAI was at its peak influence, serving as the multi-stakeholder venue for AI ethics discussions before the frontier-model era.
From 2020 through 2023 its relevance declined as several shifts converged: frontier labs built out their own internal policy and safety teams (Anthropic, OpenAI, DeepMind); advocacy groups fragmented into narrower coalitions (CAIS, FLI, AI Now); and policy venues multiplied (OECD, GPAI, the Bletchley–Seoul summits). From 2023 to 2026 PAI focused on synthetic-media transparency through its Responsible Practices for Synthetic Media framework, worker-voice work, and safe foundation-model deployment guidelines. Its policy salience remained below its 2018–2019 peak.
Activities and outputs
PAI's work spans several strands. Its Responsible Practices for Synthetic Media framework (2023) is a multi-stakeholder framework on deepfake and generative-media transparency, signed by OpenAI, TikTok, and others. Its earlier AI Incidents Database was precursor work on incident tracking that was later spun off or absorbed into other efforts. Its Safe Foundation Model Deployment guidance provides frameworks for red-teaming, evaluation, and deployment best practices. PAI also maintains worker-voice and labor research strands.
Positioning and reception
PAI operates as a multi-stakeholder convener, distinct from several adjacent organizational types. Pure trade associations such as the Frontier Model Forum are narrower industry coalitions that gained salience after 2023. Civil-society advocacy groups such as AI Now take a more critical posture. Existential-risk nonprofits such as CAIS and MIRI adopt a different framing.
Critics argue that PAI has been structurally captured by its industry funders, contending that the multi-stakeholder model produces lowest-common-denominator consensus and has been sidelined as frontier labs prefer bilateral or smaller-coalition engagements such as the Frontier Model Forum. Defenders argue that PAI's multi-stakeholder breadth remains valuable for topics such as synthetic media and worker voice where narrower coalitions will not engage.
Funding and governance
PAI is funded principally by corporate partner dues, supplemented by foundation grants. Critics identify the corporate-dues model as the source of the structural incentive they describe.
Relationships
- related: Frontier Model Forum — narrower industry successor for frontier-lab-specific coordination
- related: AI Now Institute — critical counterweight; overlapping membership in synthetic-media discussions
- related: OECD AI Principles — international governmental parallel (page pending)
- related: OpenAI, Anthropic, Google DeepMind, Microsoft — lab members with varying engagement
- related: synthetic media / deepfakes — core PAI topic area
Notes
No primary PAI source is currently in the wiki. This page is synthesized from PAI's public footprint. Confidence medium pending source ingest.