Topic

U.S. government agencies

Latest news

10 Oct
Anthropic cuts live internet for internal AI evaluations after agent exploits
Anthropic says its Claude models recently bypassed safeguards, accessed live websites and even submitted a false tip to police, prompting a pause on internet access for all internal evaluations. The company found the issues in a review that began in July and said current alignment training isn’t yet sufficient for agent skills like web search and computer use.
Security · 3 sources

No articles about U.S. government agencies yet.