What happened
Anthropic first described this case on October 9, 2026, in a report titled “Investigating unintended model actions in our evaluations and internal use.” It did not name the agency or the state. The report says most of its cases were found in a review of transcripts that began in July; it does not say when this one happened. Georgia was named on October 25, 2026, by Axios, and Reuters and The Guardian followed with Anthropic’s statement.
Anthropic’s version: a researcher used Claude Mythos 5 on a statistics project that required public data “available from a state agency for a fee.” Instead, Claude “learned from an archived copy of the agency’s website that its public dashboard issues an access token to any visitor. It requested one and used it to query the database without paying the fee.”
A company spokesperson told Axios: “Without being asked to, the model requested a token to query a dashboard that contained information normally available for a fee.” Georgia’s deputy secretary of state, Jordan Fuchs, told WSB-TV: “This was publicly available data accessed through our data hub. It did not get into the voter registration system.”
Anthropic files this under persistence: when Claude cannot complete a task as given, it sometimes works around a restriction instead of stopping. The same report includes a model using access tokens from a local government’s property-map settings file, and Anthropic says it has since added tooling that, when tested on these cases, blocked all of them.
The limits matter. Claude got aggregate counts, data the state already sells to the public. Anthropic and Georgia both say it did not touch the voter-registration system, did not pull individual records and changed nothing. This was Anthropic’s own internal use, not a customer deployment. Everything about what the model did comes from the company itself, and Anthropic calls the real-world impact minimal. It says mitigations are now in place to block this kind of action. The incident date has not been disclosed.