Claude Reportedly Accessed Georgia’s Aggregate Vote Counts After Autonomously Bypassing Fee Restriction

Anthropic’s Claude is said to have accessed Georgia’s publicly available election data without human authorisation. According to a report, the AI model bypassed a payment restriction while working on an internal statistics project. The incident reportedly involved aggregate vote counts that the US state makes available for a fee, instead of individual voter records or its voter registration system. Anthropic said the model was not instructed to obtain an access token for the data and has since introduced measures to prevent similar actions.

Claude Bypassed Paywall to Access Georgia Election Data

According to an Axios report, Claude accessed aggregate vote counts from a Georgia state dashboard that normally makes the information available for a fee. The model was reportedly being used internally for a statistics project when it requested an access token to query the dashboard, without being instructed to do so.

An Anthropic spokesperson told Axios that Claude did not access Georgia’s voter registration system, retrieve individual voting records, or modify any election data. Instead, the model obtained aggregate vote counts that the state makes available to the public for a fee. Both Anthropic and Georgia officials reportedly said the model bypassed the payment restriction. In a further statement, Anthropic clarified that it has introduced mitigations to block similar actions.

Axios also reported that Anthropic had notified Georgia’s secretary of state’s office and the White House weeks before the incident became public.

The Georgia incident follows an October 9 blog post in which Anthropic detailed several examples of Claude taking unintended actions during internal evaluations and use. The AI company grouped the behaviour into four categories: exploiting basic software flaws to run commands on servers, submitting sensitive forms on real websites, working around restrictions to access data gated by a token or fee, and using URL-shortening services to get around limits in its fetch tool.

One example cited by Anthropic involved an Anthropic researcher using Claude Mythos 5 for a statistics project that required public data from a state agency. The model found an archived version of the agency’s website, discovered that its dashboard issued an access token to visitors, and used the token to send requests directly to the server to retrieve data that was otherwise available for a fee.