Third-Party Safety Auditing of Frontier AI Models
Listen to the latest briefing
Prefer to listen? Hit play — or read the full story below.
At a glance
Third-party safety auditing concerns independent checks on powerful frontier AI models. As of October 15, 2026, has given six evaluators from nonprofit employee-like access, including company hardware, office space and internal tools comparable to its own risk assessment teams, with sharing encouraged except for customer data, privileged legal information and model architecture details. The arrangement matters because deeper external oversight could build trust in AI systems and set a precedent for how AI labs handle safety evaluation across the industry.
Story updates
2d agoLatest
Anthropic Grants Deep Access to Outside Safety Testers
On October 15, 2026, Anthropic said it will give employee-like access to six evaluators from METR, a nonprofit that tests frontier models. With no prior arrangement on record, this creates a new level of third-party oversight, with evaluators receiving company hardware, office space, and internal tools roughly on par with Anthropic's own risk teams.
Staff are urged to share information freely, except for customer data, privileged legal material, and model architecture details. The move could set a new industry bar for independent AI safety auditing and trust.
Make a prediction
Stay on top of Third-Party Safety Auditing of Frontier AI Models
You'll be notified here when this topic meaningfully develops.
Stream spatial intelligence