Anthropic’s Claude got around a paywall to access Georgia voters’ data without human authorization.
Anthropic confirmed to Axios Sunday that the company’s AI model had, without prompting, requested a digital entry pass, or “access token,” to collect aggregate vote counts for a statistics project.
“Claude did not access the state’s voter registration system, change any voting records, or pull any individual voter records,” Anthropic said, adding that it had placed guardrails to prevent the chatbot from taking these sorts of actions in the future.
The company first notified the Georgia secretary of state’s office and the White House three weeks ago about the spontaneous access. Trump administration officials urged the state office to disclose the incident, but those calls went unheeded, a source told Axios.
This revelation comes as Anthropic unveiled its public IPO filing Monday, ahead of what is expected to be one of the largest stock debuts in history at an estimated value of more than $2 trillion. It also comes weeks after one of its researchers warned that the company’s product could cause a mass extinction event.
This is just the latest instance of an artificial intelligence company being forced to disclose its model’s unauthorized behavior.