Global network refocuses on AI measurement and evaluation; UK named Network Coordinator
Published 9 December 2025
Governments need clear, comparable ways to test advanced AI. The International Network of AI Safety Institutes has formally become the International Network for Advanced AI Measurement, Evaluation and Science - a shift that puts scientific evaluation at the centre of international cooperation. The UK will coordinate the network going forward.
What changed
At its latest meeting in San Diego, the group agreed to re-double collective work on the science of measuring and evaluating advanced AI systems. The rebrand reflects a long-term focus: building methods and benchmarks that can keep pace with fast-moving model capabilities. The UK will serve as Network Coordinator, helping partners align practical approaches to testing.
Who is involved
- Australia
- Canada
- European Union
- France
- Japan
- Kenya
- Republic of Korea
- Singapore
- United Kingdom
- United States
Formed in November 2024, the network has since shared knowledge on AI evaluations to help countries align on best practices.
Why this matters for government
- Trust and adoption: Clear testing builds public confidence, which helps unlock innovation, investment, and service improvements.
- Comparable standards: Shared evaluation methods make procurement and assurance simpler across borders and agencies.
- Risk management: Rigorous testing supports proportionate controls for safety, security, and reliability before deployment.
Meeting context
The San Diego session took place alongside NeurIPS, one of the leading AI research conferences, enabling deeper exchanges with researchers and industry. Network members reaffirmed their focus on practical, testable methods, not just principles.
What leaders said
AI Minister Kanishka Narayan: "Trust in AI isn't a choice-it's a necessity. That's why the Network's mission, and our new role as its coordinator, is to make trust the foundation of progress. By bringing nations together on evaluation and research, we'll lead from the front, shape the global future of AI, and unlock its benefits for everyone."
Adam Beaumont, Interim Director of the AI Security Institute: "Advanced AI systems are being developed and deployed globally, so our approach to evaluating them has to be global too. That is why this Network is so important. Our focus will be on supporting partners to develop rigorous, practical ways of testing advanced AI."
What public sector teams can do now
- Map current and planned AI use cases, then document the evaluation criteria you will require before deployment (safety, security, fairness, reliability, and performance).
- Adopt a standard risk and testing rubric so teams speak the same language. As a reference point, see the NIST AI Risk Management Framework.
- Pilot shared test suites with partner agencies. Compare results and capture lessons learned to inform procurement and oversight.
- Set up a review cadence. Re-test AI systems on a schedule and after major model updates or context changes.
- Track outputs from the network to align your evaluation playbooks with international methods as they mature.
About the network's direction
The network's refocus is about making evaluation science actionable: clear tests, repeatable methods, and results that different countries can compare. This will help governments set requirements that vendors can meet and auditors can verify.
Media enquiries
DSIT media enquiries
Email: press@dsit.gov.uk
Phone: 020 7215 3000 (Monday to Friday, 8:30am to 6pm)
Upskilling your team
If you're planning capability-building for evaluation and assurance roles, explore role-based options at Complete AI Training - Courses by Job.
Your membership also unlocks:
