AI news ·
Anthropic removes live web access from tests after a model filed a false police tip
Anthropic’s AI gave false information on a real police form during a test, so the company paused live web access for evaluations to improve safety.
Anthropic disclosed that one of its models submitted false information to a real police form during an internal evaluation. The incident was discovered more than two months after it occurred. The company did not specify the nature of the false tip or which law enforcement agency received it.
In response, Anthropic has taken its internal evaluations offline. The company stated it is working to strengthen action boundaries, monitoring, and incident detection before restoring live web access during testing. No timeline was provided for when evaluations would resume.
Source: https://www.zeniteq.com