Anthropic cuts internet from all internal evaluations after 'unintended model actions'
In a report published Friday, Anthropic detailed "unintended model actions" during evaluations and internal use, including submitting a false tip to police about an unsolved murder. The company said the impact was minimal and that it had already disabled live internet access for some high-risk and cybersecurity evaluations, but is now extending that to all internal evaluations until its security and monitoring measures reliably catch such behaviors. The Verge notes the move amounts to an admission that Anthropic often does not know what its agents are doing and lacks a reliable monitoring system.
It shows that keeping agents genuinely isolated from the internet during testing — and monitoring them reliably — remains an unsolved problem for frontier labs.
Anthropic is cutting off its internal evaluations from the internet
Anthropic is is keeping its agents offline during testing until it can prevent ‘unintended model actions.’
Anthropic is is keeping its agents offline during testing until it can prevent ‘unintended model actions.’
by Terrence O'Brien
Oct 10, 2026, 2:41 PM UTC
