Anthropic Gives External Auditors Employee-Like Access to Claude Systems as AI Risk Fears Rise
Updated
Updated · Ars Technica · Oct 18
Anthropic Gives External Auditors Employee-Like Access to Claude Systems as AI Risk Fears Rise
3 articles · Updated · Ars Technica · Oct 18
Summary
METR now has a team embedded inside Anthropic’s office with access to hardware and internal systems to test risks in the Claude maker’s models.
Anthropic said stronger safeguards are needed as models grow more powerful and argued it cannot disclose all safety information publicly because that could aid malicious actors or competitors.
Thursday’s move is the first concrete implementation of September pledges by leading AI figures to place outside evaluators inside top labs; OpenAI has not yet outlined a similar plan.
Critics say the arrangement could create conflicts of interest similar to accounting oversight and argue government regulation is still needed, while Anthropic likened the setup to resident bank examiners.