Anthropic’s Claude artificial intelligence model bypassed a paywall to access Georgia voter data without human authorization, the company and Georgia officials said, a disclosure that set off alarm days before the midterm elections.
Claude accessed aggregate vote counts while working on an internal Anthropic statistics project, Axios reported Sunday. The data is publicly available from the state for a fee, but the model got around the paywall to reach it, according to Georgia officials and Anthropic.
“Without being asked to, the model requested a token to query a dashboard that contained information normally available for a fee,” an Anthropic spokesperson told Axios. “We have since put in place mitigations to block these kinds of actions.”
The spokesperson said Claude did not access the state’s voter registration system or change or pull any voting records.
Anthropic notified the Georgia secretary of state’s office and the White House weeks ago, according to Axios. A source told the outlet that administration officials urged Georgia to disclose the incident sooner, and that those requests went unheeded.
The disclosure comes as federal officials are already warning election officials to be on guard against AI-powered threats, and as campaigns face a wave of deepfaked attack ads.
It drew speedy reaction online, with widespread blame placed squarely on the company.
“If done by a human, this would be a federal crime,” wrote Juliette Kayyem, a former assistant secretary of Homeland Security and CNN national security analyst.
Holly Elmore, who leads the AI safety group PauseAI US, wrote that “actually Dario Amodei and the staff of Anthropic did this,” referring to the company’s CEO. An anonymous law professor account, Lawprofblawg, called for “a criminal investigation of who allowed Claude to do that.”
Republican strategist and CNN commentator Shermichael Singleton asked, “What’s going on at Anthropic?”
Zeteo founder and journalist Mehdi Hasan wrote, “Come on! FFS,” while Alyssa Farah Griffin, a former Trump White House strategic communications director, responded, “Cool, cool.”
Tyler Rogoway, editor-in-chief of the defense news site TWZ, warned that “we are going to really regret how the genesis of AI was handled,” calling AI the “most destabilizing weapon since the nuclear bomb.”
The incident is the latest in a string of disclosures about AI models taking unauthorized actions.
Last month, Axios reported that OpenAI, Anthropic and security researchers were investigating tens of thousands of incidents in which frontier models acted in ways evaluators would consider problematic, including escaping sandboxes and bypassing guardrails. OpenAI paused training on its most capable models, and Anthropic commissioned a third-party safety review.
Anthropic has had a rocky relationship with the Trump administration. A federal judge in March blocked the administration’s bid to blacklist the company as a national security risk after a dispute with the Pentagon, and in September, senior officials scrambled over fears that a preview of Anthropic’s Mythos model could be used for damaging cyberattacks.
The post Horror as rogue AI smashes through paywall to grab voter data: ‘Would be a federal crime’ appeared first on Raw Story.
