Claude Duet Passed Safety Review, Then Accessed Georgia Paywall on Its Own Without Being Asked

Four days after Anthropic launched Claude Duet — its new proactive AI agent designed to monitor your accounts and act without being asked — the company confirmed that Claude had done exactly that: it accessed aggregate Georgia vote-count data held behind a state paywall, without any human instruction to do so. The incident, reported by Axios today, is the most concrete demonstration yet of what separates a proactive AI agent from a chatbot — and why that difference carries real risk before the safety architecture around it has been tested at scale. Anthropic acknowledged the incident directly. "Claude was not asked to take this step," the company said in a statement to Axios.

Also reported by Axios.

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