Partnership on AI

What it is

The Partnership on AI (PAI) is a nonprofit coalition created in 2016 and committed to the responsible use of artificial intelligence^[1]. It describes itself as a non-profit partnership of academic, civil society, industry, and media organizations creating solutions so that AI advances positive outcomes for people and society^[2]. It also states plainly what it is not: it is not a trade group or an advocacy organization^[2]. That distinction shapes how it works. Rather than lobbying for a single industry position, it presents itself as a place where organizations with different and sometimes opposing interests — competing companies, researchers who study them, and groups that criticize them — produce shared resources together.

Where it came from

PAI was publicly announced on September 28, 2016. Its founding members were Amazon, Facebook, Google, DeepMind, Microsoft and IBM^[1]. All of the original members were large technology companies, which shaped how the new body was received. At the announcement, the interim co-chairs were Eric Horvitz of Microsoft Research and Mustafa Suleyman of DeepMind^[1].

Membership and leadership over time

The coalition broadened steadily after launch. In January 2017 Tom Gruber, Apple's head of advanced development for Siri, joined the board^[1]. In October 2017 Terah Lyons became the organization's founding executive director^[1]. In October 2018 Baidu became the first firm from China to join the Partnership^[1]. On October 26, 2021 Rebecca Finlay was named CEO^[1].

What it works on

PAI's programs cover AI and media integrity; AI, work and the economy; justice, transparency and accountability; inclusive research and design; and security for AI^[1]. Two concrete outputs are frequently cited. In November 2020 the Partnership announced the AI Incident Database, a tool to identify, assess, manage and communicate AI risk and harm^[1]. In February 2023 it launched a framework guiding the ethical development and use of synthetic media, backed by initial partners including Adobe, BBC, CBC/Radio-Canada, Bumble, OpenAI, TikTok, WITNESS, Synthesia, D-ID and Respeecher^[1].

Scale and strategy

In 2024 the Partnership reported that it included 126 partner organizations from academia, civil society and industry based in 16 countries^[1]. Its strategic plan, "Shared Stakes, Lasting Impact", lays out how PAI will spend the next five years building trust across the AI value chain, anchored in two focus areas: AI and human connection, and AI and shared prosperity^[2].

Where it sits in AI governance

Debate about AI involves parties with very different roles: companies that compete with one another, researchers who study their systems, and civil society groups that criticize them. An organization that convenes all of these can do something none of them does alone — agree on definitions, share vocabulary, and build common measuring tools. An incident database is a clear example: a shared record of AI harms is useful precisely because it is not compiled by a single vendor, and no single company has much incentive to compile it. The same logic applies to frameworks for synthetic media, which only work if the platforms producing such media and the organizations detecting and reporting on it use the same terms.

The coalition form also sets limits. Because PAI is not an advocacy organization, its output tends to be frameworks, databases and recommendations rather than binding rules; it does not set policy for anyone. And a body whose founding members were the companies it examines always carries a tension between access and independence, which is part of why its membership and leadership have broadened over the years to include academia, civil society and media.