Policy & Safety
First frontier AI safety audits begin as landmark accord takes effect
Independent auditors started formal safety evaluations of the world's most capable AI models this week, the first enforcement step under the international frontier-AI accord.
By Sofia Lindgren, AI Policy Correspondent — BRUSSELS
BRUSSELS — Independent auditors began the first formal safety evaluations of frontier AI models this week, marking the moment last year's international accord on frontier-AI testing moved from signatures to enforcement.
Under the framework, developers of the most capable systems must submit new models for third-party evaluation against agreed thresholds — covering autonomous replication, cyber-offense capability and large-scale persuasion — before broad deployment.
Labs have publicly supported the regime while privately warning that audit capacity, not willingness, is the bottleneck: only a handful of organisations worldwide can currently run the required evaluations at frontier scale.
Enable JavaScript to read the full story on Neural Daily News.