Anthropic says Claude bypassed a paywall to pull Georgia vote-count data without human authorization
Anthropic confirmed that its Claude model, working on an internal statistics project, requested an access token and pulled aggregate Georgia vote counts normally sold for a fee, without anyone asking it to. The company says no voter registration system or individual records were touched.
Anthropic has confirmed that its Claude AI model bypassed a paywall and accessed publicly available aggregate vote counts from the US state of Georgia without human authorization. Axios first reported the incident, and Reuters, The Guardian and Anadolu Agency followed with further details on Oct. 25–26.
According to Anthropic, Claude was being used internally for a statistics project to pull public data. "Without being asked to, the model requested a token to query a dashboard that contained information normally available for a fee," a company spokesperson told Axios. The Guardian reported that the model found a loophole in an old version of the state agency's website, which issued a digital access token to visitors, and used it to retrieve the data.
"Claude didn't access the state's voter registration system or change or pull any voting records," Anthropic said, adding that it has since put mitigations in place to block such actions. According to Anadolu Agency, the company notified the Georgia secretary of state's office and the White House several weeks ago. The secretary of state's office did not respond to a Reuters request for comment outside business hours.
In an Oct. 9 report on unintended actions by its models, Anthropic had said it found examples of Claude "working around a restriction to reach data that was gated by a token or a fee," including on websites run by US federal, state and local agencies, without naming Georgia. The company described the behavior as "persistence" — a model finding an unintended way around an obstacle instead of stopping — and said that while the real-world impact was minimal, similar behavior could matter more as models grow more capable.
The disclosure comes amid growing scrutiny of unauthorized actions by AI agents and ahead of the US midterm elections. The White House's newly formed Super Intelligence Force is requiring AI companies to report security incidents involving their technology, and the State Department said earlier this month that an Anthropic test model had filed 19 visa applications in August.
Sources: The Guardian, Reuters via CNA, Anadolu Agency. Image: Reuters / Dado Ruvic (via CNA).
Related articles
