The International AI Safety Institute Network (INSAI, also referred to as AISIN) is an intergovernmental, non-treaty coordination network of national AI Safety Institutes and equivalent bodies. It was announced in November 2024 at the San Francisco AI Safety Institute Convening and expanded through 2025 and 2026. It serves as the multilateral technical-evaluation arm of the AI governance architecture developed through the Bletchley, Seoul, and Paris summit sequence.
| Field | Value |
|---|---|
| Type | Intergovernmental coordination network (non-treaty) |
| Established | November 2024 (announced at the San Francisco AI Safety Institute Convening) |
| Members (as of 2025–2026) | 11+ AI Safety Institutes and equivalents including the US, UK, Japan, Singapore, Canada, France, Kenya, Australia, South Korea, the EU (AI Office), and others; membership has expanded through 2026 |
| Secretariat / convening rotation | Originally US-led (San Francisco convening); rotating host model emerging through the Bletchley–Seoul–Paris summit sequence |
Origin and purpose
The network emerged from the Bletchley–Seoul–Paris summit sequence (November 2023, May 2024, February 2025) as the technical implementation arm of the political commitments in those declarations. The Bletchley Declaration (November 2023) committed signatories to shared concern about frontier AI risks and to cooperation on evaluation. The AI Safety Institute model was established with the UK AISI (November 2023) and the US AISI (November 2023, housed at NIST) as the two originals, with other jurisdictions following. The Seoul Commitments (May 2024) formalized government roles in reviewing frontier lab safety frameworks. The San Francisco Convening (November 2024) formally launched the network.
At the Paris AI Action Summit (February 2025), the US and UK did not sign the Paris declaration, but the AISI network continued operational cooperation.
Functions
As of 2026, the network's operational work includes joint pre-deployment evaluations of frontier models, in cooperation with labs; shared evaluation methodology and test-set development, building on open-sourced tools such as the UK AISI's Inspect framework; information sharing on capability trends, safety incidents, and red-teaming results; coordination on red-lines and thresholds, though formal threshold agreement remains aspirational; and talent and research exchange among member AISIs.
Member institutes
Membership has continued to expand, and specific lists shift. Representative members include:
| AISI / Equivalent | Jurisdiction | Status | |
|---|---|---|---|
| [[us-ai-safety-institute | US AISI]] | US (NIST) | Founding member; continuity questions under post-EO-14110 policy |
| [[uk-ai-safety-institute | UK AISI / AI Security Institute]] | UK (DSIT) | Founding member; renamed Feb 2025 |
| Japan AISI | Japan (IPA) | Member | |
| Singapore (DTC / IMDA-adjacent) | Singapore | Member | |
| Canada AISI | Canada | Announced 2024 | |
| France (INESIA-adjacent) | France | Member | |
| EU AI Office | EU | Participates; functions via the EU AI Act | |
| South Korea AISI | South Korea | Member | |
| Kenya | Kenya | Member (announced) | |
| Australia | Australia | Voluntary AI safety standard route; participates |
Key outputs and reports
The International AI Safety Report (Bengio et al., 2025) is the scientific consensus document associated with the network, modeled on IPCC reports. Member institutes have published individually, including UK AISI Frontier AI Trends Report 2025, UK AISI — Advanced AI Evaluations May Update, and US AI Safety Institute — Vision, Mission, and Strategic Goals. The network also produces joint technical notes on evaluation methodology.
Relationships
- depends-on: The Bletchley Declaration (AI Safety Summit, 1–2 November 2023), Frontier AI Safety Commitments (Seoul, 2024)
- related: US AI Safety Institute (NIST AISI), UK AI Safety Institute (AI Security Institute) — founding institutes
- related: International AI Safety Report 2025 — associated scientific report
- related: UK AISI Frontier AI Trends Report 2025, UK AISI — Advanced AI Evaluations May Update, US AI Safety Institute — Vision, Mission, and Strategic Goals
- related: Apollo Research, METR — independent evaluators whose work connects to AISI evaluations
- related: Paris AI Action Summit Declaration (2025) — summit context
- related: AI Safety Cases and Frameworks, AI Benchmarks and Evaluation
- related: Frontier Model Forum — industry-side counterpart