Centre for the Governance of AI
AI-generated text. This page was generated using artificial intelligence.
Centre for the Governance of AI (GovAI) is a think tank studying how governments and other institutions can respond to advanced artificial intelligence. It combines public research, analysis for decision-makers and programs intended to develop expertise in AI governance. Its research includes risk management, economic effects, frontier regulation, geopolitics and technical approaches to oversight.[1]
Institutional development
GovAI's institutional history traces its beginnings to Yale University in 2016 under Allan Dafoe, its establishment as an Oxford academic center in 2018, and its departure from the university in 2021 to become an independent nonprofit. Its current organizational disclosure identifies a US section 501(c)(3) entity, EIN 99-4000294, and a UK subsidiary.[1]
GovAI reports engagement with policymakers, research programs and corporate safety frameworks. That account does not establish causal influence over government decisions. Its publication model also matters: research blog posts identify their arguments as those of their authors, rather than institutional positions.[1][2]
Compute governance
A February 2024 research summary by Lennart Heim and colleagues examines computing resources as a point of intervention in AI development. The authors distinguish three functions: monitoring activity to improve visibility; allocating access through measures such as subsidies or restrictions; and using hardware or infrastructure to enforce rules. They argue that physical chips and a concentrated supply chain can be easier to monitor than widely transferable algorithms.[3]
The same analysis emphasizes risks. Monitoring can expose private information, centralized control can strengthen powerful firms or governments, and poorly targeted restrictions can burden beneficial research. The authors recommend attention to scope, privacy and procedural safeguards. The underlying collaboration included researchers from academia, civil society and industry, and the publication cautions that coauthorship does not imply agreement with every statement or endorsement by each author's employer.[3]
Research on safety practices
In June 2023, Jonas Schuett and colleagues reported a survey of 51 experts from AI laboratories, academia and civil society. Respondents assessed 50 proposed safety and governance practices. The study found particularly strong support for pre-deployment risk assessment, dangerous-capability evaluation, third-party audits, safety restrictions and red teaming. Its contribution was to document the views of a defined expert sample about practices laboratories should adopt.[2]
The results should not be read as a public opinion poll, a finding that laboratories had implemented those practices, or a legal standard. The researchers reported 51 responses to 92 invitations and described the questions and response scale. That methodology helps readers assess what the reported agreement establishes and where its reach is limited.[2]
Infrastructure for AI agents
Alan Chan's July 2025 analysis considers who should supply protocols that allow agents to act safely and interact with digital services. Examples include limited permissions, payment authorization, identity verification and mechanisms for reversing unwanted actions. Chan argues that businesses have incentives to build infrastructure that makes their own products more useful and controllable.[4]
He identifies a weaker market incentive to address systemic risks that users may not consider when choosing a product. The analysis also raises questions about security, interoperability and equal treatment of agents. It therefore leaves a role for public institutions even when private provision is likely. This is a proposed allocation of responsibilities, not a claim that government must operate every component of agent infrastructure.[4]
Selected contributions
- Towards Best Practices in AGI Safety and Governance survey summary (2023): expert views on laboratory governance.[2]
- Computing Power and the Governance of AI (2024): opportunities and safeguards for compute-based intervention.[3]
- What Role Should Governments Play in Providing AI Agent Infrastructure? (2025): incentives and gaps in private infrastructure provision.[4]
Related articles
References
- ↑ 1.0 1.1 1.2 1.3 1.4 1.5 GovAI, About Us. Accessed September 12, 2026.
- ↑ 2.0 2.1 2.2 2.3 Jonas Schuett and colleagues, New Survey: Broad Expert Consensus for Many AGI Safety and Governance Practices, June 5, 2023. Accessed September 12, 2026.
- ↑ 3.0 3.1 3.2 Lennart Heim and colleagues, Computing Power and the Governance of AI, February 14, 2024. Accessed September 12, 2026.
- ↑ 4.0 4.1 4.2 Alan Chan, What Role Should Governments Play in Providing AI Agent Infrastructure?, July 1, 2025. Accessed September 12, 2026.