Written and maintained by CASRAI Editorial Board
Last updated
The short answer
Partnership on AI (PAI) is a nonprofit organisation, founded in 2016, that convenes industry, academic, civil-society, and media organisations to work on the responsible development and use of AI. Its own materials describe it as “a non-profit partnership of academic, civil society, industry, and media organisations” and state explicitly that it is “not a trade group or advocacy organisation.” That multi-stakeholder membership — including non-industry organisations as full partners, not just external advisors — is the main thing that distinguishes PAI from industry-only coordination bodies such as the Frontier Model Forum, whose formal membership is limited to frontier AI companies (Amazon, Anthropic, Google, Meta, Microsoft, and OpenAI).
Founding and history
PAI was publicly announced on 28 September 2016, with founding members Amazon, Facebook (now Meta), Google, DeepMind, Microsoft, and IBM. Apple joined a few months later, in January 2017, through its Siri head of advanced development, Tom Gruber. Terah Lyons became the organisation’s founding executive director in October 2017. Baidu joined in October 2018, becoming the first company based in China to join the Partnership.
Membership has expanded well beyond the founding technology companies since then. By 2024, PAI reported roughly 126 partner organisations spanning academia, civil society, and industry, based in 16 countries; PAI’s own site now cites a partner network of 150+ organisations, including academic institutions (among them MIT, Stanford, UC Berkeley, the Oxford Internet Institute, and the Alan Turing Institute), civil-society groups, media organisations, and UN agencies, alongside the technology companies. Not every early relationship held: Access Now resigned from PAI in October 2020, saying the organisation had “an increasingly smaller role for civil society” and citing disagreements with its approach to AI ethics and risk assessment. Rebecca Finlay has served as PAI’s CEO since October 2021.
What PAI actually does
PAI’s stated mission is “bringing diverse voices together across global sectors, disciplines, and demographics so developments in AI advance positive outcomes for people and society.” In practice, that work is organised around a small number of ongoing programme areas, per PAI’s own site:
- AI Safety — work on frontier-AI risk questions, sitting alongside PAI’s older, broader ethics work rather than replacing it.
- AI, Labor, and the Economy — the effects of AI deployment on jobs and economic structures.
- AI & Human Connection and Enterprise AI — applied and social-impact work.
- Public Policy and a European Steering Committee — engagement with regulators and policymakers.
- Philanthropy — grant-funded initiatives.
Two concrete, dated outputs are worth naming directly. In November 2020, PAI launched the AI Incident Database (AIID), a public, searchable record of real-world harms and near-misses involving deployed AI systems — one of PAI’s most widely cited artefacts outside AI-policy circles. In February 2023, PAI published a framework for the ethical development and deployment of synthetic media, addressing AI-generated audio, video, and images.
How PAI differs from the Frontier Model Forum
Both organisations sit in the frontier-AI-governance space, and both are sometimes mentioned in the same breath, but they are structured differently:
- Membership. PAI’s partners include academic institutions, civil-society organisations, and media outlets as full members, alongside industry. The Frontier Model Forum’s formal membership is limited to a small set of frontier AI developers — its own site describes working “across academia, civil society, industry and government” but that describes external collaboration, not membership.
- Scope and age. PAI is the older organisation by several years (2016 vs. the Forum’s 2023 founding) and its programme areas span AI ethics, labour, and policy broadly, not only frontier-model safety specifically.
- Governance model. PAI positions itself explicitly as not a trade group; the Frontier Model Forum is a coalition of the companies that build frontier models, organised around technical safety standards for that class of system.
The two are not competitors so much as different answers to “who should be at the table”: PAI’s answer includes civil society and academia as members by design; the Forum’s answer, for formal membership, does not.
Where this fits in AI governance
PAI is one of several non-governmental bodies that have emerged to coordinate AI-safety and AI-ethics work outside formal regulation — alongside industry-only groups like the Frontier Model Forum and government bodies like AI safety institutes. For a structured way to track which frameworks, evaluations, and standards apply to a given AI system or deployment, see CASRAI’s NIKOLAI.
Is Partnership on AI a regulator?
No. PAI has no enforcement power and issues no binding rules. It is a membership nonprofit that publishes research, convenes its partner organisations, and maintains projects like the AI Incident Database — influence through publication and convening, not regulation.
Is Partnership on AI the same as the Frontier Model Forum?
No. They are separate organisations with different founding dates (2016 vs. 2023) and different membership models — PAI’s partners include academic and civil-society organisations as full members, while the Frontier Model Forum’s formal membership is limited to frontier AI developers.
Who funds Partnership on AI?
PAI was originally supported by multi-year grants from its founding technology-company members and has since drawn on a broader base of partner and philanthropic funding as its membership grew; it does not publish a single current funding breakdown on its public-facing pages.
