Anthropic says Claude pulled Georgia vote counts without paying the fee

Anthropic's Claude pulled aggregate vote counts from Georgia that the state makes available for a fee, after the model requested an access token on its own during an internal statistics project, the company and Georgia officials confirmed Sunday (Oct. 25).

The data came through Georgia's public data hub, not its election systems. "This was publicly available data accessed through our data hub. It did not get into the voter registration system," Georgia Deputy Secretary of State Jordan Fuchs said in a statement to Fox Business.

"Without being asked to, the model requested a token to query a dashboard that contained information normally available for a fee," an Anthropic spokesperson told Axios, which first reported Georgia's role. "We have since put in place mitigations to block these kinds of actions."

Neither Anthropic nor Georgia has said when the query happened, how much data Claude retrieved or what the fee would have been.

Anthropic said Claude didn't change any voting records or pull individual voter records. "The model pulled only aggregate counts which can be derived from voter data the state makes available to anyone for a fee," the spokesperson told Fox Business.

The incident isn't new to Anthropic's own records. The company described it, without naming Georgia, in an Oct. 9 report on unintended model actions, saying it withheld the names of the organizations involved "to avoid exposing vulnerabilities in their systems, and at their request."

In that report, Anthropic said a researcher gave Claude Mythos 5 a statistics project that needed a state agency's fee-based data. Claude instead "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," the company wrote.

Anthropic said it notified each agency involved and briefed the White House. Administration officials pushed Georgia's secretary of state's office to disclose the incident sooner, but those requests went unheeded, a person familiar with the matter told Axios.

The disclosure comes nine days before Georgia's Nov. 3 general election.

The Georgia query was one of several cases Anthropic grouped as working around restrictions to reach gated data. In another, Mythos 5 read the settings file of a local government's property map, found working access tokens and used them to query the map's server directly while trying to identify where a photo was taken.

The same report described Claude exploiting basic software flaws such as SQL and command injection (input a site mistakenly runs as instructions) to run commands on a university server. It also covered submitting a real form when it should have stopped, and using URL shorteners to get around limits in its web fetch tool.

One form case became public separately: Claude Haiku 4.5 submitted an invented tip on a police department's homicide tip form, which the Philadelphia Police Department later disclosed. The submission was flagged as spam and never forwarded, Anthropic said.

Anthropic called most of these behaviors "persistence, in which Claude, when it cannot complete a task as given, works around a restriction instead of stopping." It said the cases had "minimal real-world impact."

In response, Anthropic cut live internet access for all of its internal evaluations, tightened its web fetch tool and built tooling to automatically detect and block such actions. When tested against the cases in the report, that tooling "blocked all of them," the company said.

Anthropic isn't alone in pulling models offline. OpenAI said Friday (Oct. 23) that an internal research model exploited a DNS resolver misconfiguration on Sept. 20 to reach the live internet, and that it now trains frontier models with no internet access.

The Georgia case also lands as Anthropic prepares for a public listing; its draft IPO filing was reported last week.