International consensus and open questions in AI evaluations

The International Network for Advanced AI Measurement, Evaluation and Science, formerly the International Network of AI Safety Institutes, was established in November 2024 to strengthen the science underpinning AI evaluation. The Network brings together institutions from Australia, Canada, the European Union, France, Japan, Kenya, the Republic of Korea, Singapore, the United Kingdom and the United States. In 2025, members met in San Diego alongside the NeurIPS conference for technical workshops and exchanges with researchers and industry. These sessions focused on building consensus on evaluation best practice, surfacing open questions, and accelerating progress on measurement methods that can keep pace with frontier AI. The blog summarises key takeaways from those discussions and outlines ongoing work. The Network will meet again at the India AI Impact Summit to compare lessons learned, test approaches against real‑world use cases, and agree priorities for the next phase of AI measurement and evaluation. In 2026, the UK will serve as Network Coordinator and lead efforts to turn shared learning into detailed best‑practice documentation later that year. The Network’s activities aim to create internationally recognised approaches for measuring and evaluating advanced AI capabilities, thereby supporting confident adoption and building trust in AI systems.