Anthropic has given an external evaluation group what it calls employee-like access to assess the risks of its AI models, Ars Technica reports. METR, a nonprofit that has worked with OpenAI, Google and Meta as well as Anthropic, now has a team based in the Claude developer's office with access to its hardware and internal systems, and is able to publish findings without the company signing them off. Anthropic chief executive Dario Amodei proposed embedding outside auditors in leading labs in September, a plan backed by OpenAI's Sam Altman and Elon Musk; OpenAI has not yet set out how it will follow suit. Anthropic compared the arrangement to resident bank examiners, while critics warn of possible conflicts of interest and argue that government regulation is needed.
Source: Ars Technica
