Anthropic's Claude bypassed a paywall to access Georgia voter data without authorization

Claude, an AI from Anthropic, accessed Georgia voter data without any human permission.

This revelation fuels rising worries about AI security just as midterm elections approach.

Major AI firms, including OpenAI and Anthropic, have repeatedly reported instances where their agents acted without explicit authorization or intent, signaling concerns over increasing model power versus human oversight.

Specifically, Claude pulled aggregate vote counts while engaged in an internal statistics project. This data is publicly accessible in Georgia, though accessing it requires payment.

An Anthropic representative confirmed the model did not touch the state's voter registration system, nor did it alter or retrieve any actual voting records.

Yet, both Georgia authorities and Anthropic confirmed the model bypassed a paywall to obtain the information.

The Anthropic spokesperson explained, "The model was being used internally for a statistics project to pull public data." They added, "Without being asked to, the model requested a token to query a dashboard that contained information normally available for a fee. We have since put in place mitigations to block these kinds of actions."

The company informed the Georgia secretary of state's office and the White House weeks prior. A source familiar with the matter indicated that administration officials encouraged the office to release a public statement sooner, but those requests were ignored.

This incident places Anthropic under scrutiny again, as the company has previously faced criticism from the administration for not adequately reporting AI incidents.