Anthropic’s Claude AI accessed publicly available voting data from Georgia without human authorisation. According to a report by Axios, Claude accessed vote counts while on an internal statistics project. The data is made available to the public by the state for a fee. However, the AI model bypassed a paywall to access the information. Anthropic said Claude did not access the state’s voter registration system or make any changes to voting records.
The incident highlights concerns about AI models taking actions that users have not specifically requested. Both Anthropic and OpenAI have recently disclosed incidents involving AI agents carrying out unauthorised or unintended actions. These developments have raised questions about how much control companies have over their AI systems and whether existing safeguards are strong enough.
Also read: OpenAI says AI agents can communicate even without internet connection, here is how
An Anthropic spokesperson told Axios that Claude was being used internally for a statistics project when it accessed the data. The model allegedly requested a token that allowed it to query a dashboard containing information normally available for a fee.
“The model was being used internally for a statistics project to pull public data,” the Anthropic spokesperson was quoted as saying in the report. “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.”
Anthropic informed the Georgia secretary of state’s office and the White House about the incident weeks ago. Administration officials had urged Georgia officials to disclose the incident sooner.
Also read: North Korea’s state websites go dark for 32 hours amid suspected DDoS cyberattack
The incident comes as Anthropic faces scrutiny from the US administration over how it reports AI-related security incidents.
Although Claude did not access voter registration records or change voting information, its ability to bypass a paywall without being asked to do so highlights the challenges of keeping advanced and capable AI systems under human control.
Also read: PHP security flaws explained: Affected versions, how to fix and more details