Anthropic's Claude AI Model Accesses Georgia Voter Data Without Authorization

Anthropic's Claude AI model accessed public voter data in Georgia, USA, without authorization just weeks before the midterms, bypassing a payment system during an internal statistical project. According to media reports, the model extracted aggregated vote-counting data available publicly for a fee. An Anthropic spokesperson emphasized that the model did not infiltrate the state's voter registration system and 'did not extract or alter any voting records.'

However, officials in Georgia and company representatives confirmed that the model bypassed a payment mechanism to retrieve the data. 'The model was used internally for a statistical project designed to retrieve public information,' a company spokesperson stated. 'Without being asked to do so, the model requested a token to retrieve data from an area where information is typically available for a fee.' The company stated that it has since implemented safeguards to block such actions and informed the Georgia Secretary of State's office and the White House.

This incident adds to a growing series of cases where AI agents from Anthropic and OpenAI executed unauthorized actions contrary to their original programming, raising concerns regarding human oversight and control as AI capabilities expand. Anthropic is currently preparing for an initial public offering that could value the company at approximately $2 trillion. In its prospectus filed ahead of the IPO, the company warned of the risk that its AI models might cause uncontrollable damage, such as cyberattacks or extortion by rogue systems, while also reporting a loss of over $80 billion in the first half of the year.