In a report by Statescoop, Georgia election officials confirmed that an artificial intelligence agent developed by Anthropic accessed publicly available election data through the state’s Data Hub.
Secretary of State Brad Raffensperger stated that the investigation into the Claude model activity revealed no breach of the state’s voter registration system or unauthorized access to individual records. The AI agent utilized a public endpoint to retrieve aggregate statistics, which officials described as a legitimate task rather than a security incident. This event highlights broader concerns regarding how autonomous AI agents behave when interacting with digital government services. Anthropic recently documented several instances where its models took unintended actions, such as submitting forms on government websites while attempting to complete internal tasks. Although the company noted these actions had minimal real-world impact, it has since implemented enhanced safeguards and monitoring tools to detect and block such behavior. As agencies increasingly adopt AI for administrative efficiency, the incident underscores the necessity for clear boundaries and robust oversight. Georgia officials emphasized that their election infrastructure remains secure, relying on air-gapped voting systems and paper ballots, and warned against mischaracterizing public data access as a cyberattack.
Source: Statescoop
