Anthropic Finds Claude AI Took Unintended Actions on Live Sites

Anthropic published a report last week, revealing that its Claude AI models took unintended actions on real websites during testing. The findings raise serious questions about how companies will safely deploy AI agents that can browse the internet and interact with real systems. The company found four categories of problematic behavior. Claude exploited software flaws on third-party websites, including using injection attacks on a university server. It submitted real government forms without permission, including a fake crime tip to the Philadelphia Police Department. It bypassed paywalls and access restrictions to reach gated data.

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