Government AI Safety Organizations (Overview)

Overview

Governments have begun establishing dedicated institutions to address AI safety risks, with AI Safety Institutes (AISIs) emerging as a key organizational model since 2023. These bodies operate with public mandates and budgets that distinguish them from the largely philanthropically-funded nonprofit landscape, though they face different constraints including political cycles and bureaucratic processes.

National AI Safety Institutes

The UK AISI was the first national AI Safety Institute, established following the Bletchley Park AI Safety Summit in November 2023. The US AISI was established within NIST in 2024. Both conduct pre-deployment evaluations of frontier AI models.

Intergovernmental Bodies

• Global Partnership on Artificial Intelligence (GPAI)OrganizationGlobal Partnership on Artificial Intelligence (GPAI)GPAI represents the first major multilateral AI governance initiative but operates as a non-binding policy laboratory with limited enforcement power and structural coordination challenges. While pr...Quality: 50/100: Multilateral initiative with 29 member countries working on responsible AI development and governance

International Network of AI Safety Institutes

As of early 2026, 11+ countries have established or announced AI Safety Institutes, forming a growing network for international safety coordination. Members share evaluation methodologies, coordinate on frontier model assessments, and develop common safety benchmarks. Key members include the UK, US, Japan, Canada, France, and South Korea, with India joining in 2026.

Key Dynamics

Political vulnerability: Government AI safety bodies are subject to changes in political leadership and priorities. The US AISI's mandate and funding depend on congressional and executive support, which can shift between administrations.

Relationship with labs: AISIs must balance cooperative relationships with frontier labs (needed for access to models) against independent oversight mandates. The UK AISI has voluntary agreements with major labs for pre-deployment access.

Complementarity with nonprofits: Government bodies focus on standards, regulation, and institutional evaluation capacity, while nonprofit safety organizations (like METROrganizationMETRMETR conducts pre-deployment dangerous capability evaluations for frontier AI labs (OpenAI, Anthropic, Google DeepMind), testing autonomous replication, cybersecurity, CBRN, and manipulation capabi...Quality: 66/100 and Apollo ResearchOrganizationApollo ResearchApollo Research demonstrated in December 2024 that all six tested frontier models (including o1, Claude 3.5 Sonnet, Gemini 1.5 Pro) engage in scheming behaviors, with o1 maintaining deception in ov...Quality: 58/100) conduct more specialized technical research. There is increasing collaboration between the two.