Centre for the Governance of AI

Papers examine how rules aimed at the most capable models would affect both risk and benefit, and what trade-offs they carry. Recent work looks at auditing, embedded assessments, and whether existing rules need amending. Useful when you need a set of options rather than one recommendation.

Researchers test the evidence behind claims about cyber, bio, and misuse risks, and ask which future capabilities would change the picture. The point is to separate what is known from what is assumed. Policy teams use this to avoid overreacting to one demo or dismissing a real trend.

Work on labor markets, growth, and competition asks what governments can measure and which policy levers exist. It covers worker adaptability and the data agencies would need to track displacement. Good background for employment and competition debates that often run on guesswork.

This strand looks at technical fixes that make policy enforceable, such as identifiers for AI systems and visibility into agents. It sits between engineering and law, and it is where GovAI's research gets most concrete. Teams building compliance tooling will find the most direct use here.

GovAI runs an AI fellowship program that supports more than 100 participants a year, plus visiting programs for people already in the field. Fellows get funding, mentorship, and a peer group, and many move into government, labs, or nonprofits afterward. Admission is selective and tied to a research proposal.

Staff publish analysis of AI safety policy changes, including company scaling rules, and explain what changed and why it matters. Annual reports document funding, staff, and output in one place. That makes the organization easier to assess than a think tank that only posts op-eds.