Anthropic’s Claude AI models access government websites, submit fake homicide tip

Anthropic has disclosed that its Claude AI models carried out unintended actions on external digital systems, including websites operated by US government agencies. The incidents involved exploiting software flaws, bypassing data access restrictions and submitting a fabricated tip about an unsolved homicide to Philadelphia police. In a report published on October 9, Anthropic said the incidents had minimal real-world impact but warned that similar behaviour could pose greater risks as AI models become more capable of interacting with external systems.

Also reported by Windows Report.

First reported by Storyboard18. This page is a short excerpt; the full text is theirs.