Regulatory Consulting at Partnership on AI

Overview

Partnership on AI (PAI) is a nonprofit community of academic, civil-society, industry, and media organizations formed to address the hardest questions about AI’s future, and it functions as the field’s most established multi-stakeholder convener. Led by CEO Rebecca Finlay, PAI’s method is consensus-building: it brings together parties who disagree — frontier labs, civil-society watchdogs, academics, and policymakers — to produce shared frameworks that no single actor could impose. Its signature contribution is the Guidance for Safe Foundation Model Deployment, the first comprehensive multistakeholder guidance for foundation- and frontier-model providers, developed through a year-long series of workshops with experts affiliated with institutions including Anthropic, Google DeepMind, IBM, the Centre for the Governance of AI, and the Berkeley Center for Long-Term Cybersecurity, then opened to public comment. PAI has since built a measurement layer on top of that guidance, publishing annual transparency reporting that tracks whether model developers actually disclose the impacts of their systems. Its influence extends into international governance: Finlay co-chaired a thematic panel on safe, secure, and trustworthy AI at the UN’s first Global Dialogue on AI Governance in Geneva in July 2026, and has participated in UN initiatives including the Global Digital Compact.

AI-Focused Regulatory Work

PAI’s bet is that in a field where no one actor has legitimacy to set the rules, collectively built guardrails are worth more than unilateral ones — and that the only way to know if they work is to measure. Its 2026 Transparency Report on Foundation Model Impacts does exactly that: reviewing more than 150 papers, articles, websites, and reports to assess how well 13 organizations publicly document their models’ impacts, building on the 2023 Guidance and a 2025 progress-tracking methodology. The findings are pointed. A handful of ‘practice leaders’ are genuinely defining what to share and how, but the rest of the field lags badly, leaving a scattered and incomplete public picture. The public now has more information than ever about AI usage and potential labor-market effects, PAI notes, yet still cannot see the prevalence of harms, serious incidents, or broader societal impacts — a gap with obvious implications for anyone trying to regulate on evidence. Finlay has carried that argument onto the largest possible stage: at the UN’s first Global Dialogue on AI Governance in Geneva, which drew more than 4,000 participants, she pressed three points — the need for an open, inclusive, independent evidence base coordinated across international assessments; interoperability through shared frameworks and common standards grounded in universal principles like human rights; and the need to define and measure what ‘good AI’ practice actually looks like so progress can be shown to serve the public interest. For a directory reader, PAI is the institution to watch on whether voluntary, multistakeholder governance can produce verifiable accountability rather than well-intentioned paperwork.

Notable AI Policy Work

• Developed the Guidance for Safe Foundation Model Deployment — the first comprehensive multistakeholder guidance for foundation/frontier model providers, built over a year of workshops and released for public comment ahead of the UK AI Safety Summit.
• Published the 2026 Transparency Report on Foundation Model Impacts, measuring how 13 organizations document model impacts across a review of 150+ papers, articles, and reports — and finding the field’s disclosure landscape scattered and incomplete.
• Built a 2025 progress-tracking methodology to assess adoption of information-sharing practices among model developers, creating a repeatable accountability measure.
• CEO Rebecca Finlay co-chaired a thematic panel on safe, secure, and trustworthy AI at the UN’s first Global Dialogue on AI Governance (Geneva, July 2026) and has participated in UN initiatives including the Global Digital Compact.
• Convenes cross-sector governance efforts including the AI & Philanthropy Steering Committee (with the Ada Lovelace Institute, AI Now, Ford Foundation, Omidyar Network, and others) and joined the AI Alliance alongside 50+ institutions.

Thought Leadership

1. 2026 Transparency Report on Foundation Model Impacts
   Measures how 13 organizations publicly document foundation-model impacts across 150+ sources, finding practice leaders emerging while the rest of the field lags on information sharing.
2. Guidance for Safe Foundation Model Deployment
   PAI’s landmark multistakeholder framework giving foundation- and frontier-model providers tools for responsible development, deployment, and open-release decisions.
3. AI Will Affect Everyone’s Future. Its Governance Can’t Belong to a Few.
   Rebecca Finlay’s argument for inclusive, broadly legitimate AI governance, written around the UN’s first Global Dialogue on AI Governance.
4. Partnership on AI — resources & research
   PAI’s hub for multistakeholder work on AI safety, governance, labor impacts, and responsible deployment.

Clients

A nonprofit multistakeholder community rather than a client-serving firm. Partners span frontier AI labs and technology companies, civil-society organizations, academic institutes, and media organizations — including contributors affiliated with Anthropic, Google DeepMind, IBM, Meta, Microsoft, the Centre for the Governance of AI, and the Berkeley Center for Long-Term Cybersecurity. Engages the UN system on international AI governance.