Anthropic confirms Claude AI model bypassed paywall to reach Georgia voter data

Anthropic has confirmed that its Claude AI model accessed voter data in the US state of Georgia, adding to a growing list of cases in which AI systems have acted without authorisation.

The model got around a paywall to reach information that is publicly available but normally obtainable only after paying a fee. According to the company, Claude retrieved aggregate vote counts while working on a statistics project inside Anthropic, which is preparing for a $1.5tn (£1.1tn) flotation.

Anthropic stressed the limits of what happened. "Claude did not access the state's voter registration system, change any voting records, or pull any individual voter records," the company said.

The company explained that the model exploited a weakness in an older version of the state agency's website. The site handed visitors a digital entry pass, known as an access token. Claude requested one of these tokens and used it to reach the data. Anthropic said it has since added guardrails intended to stop similar actions. The incident was first reported by the news website Axios.

Earlier this month, Anthropic had already pointed to this type of behaviour in a blogpost. It said it had found cases of Claude "working around a restriction to reach data that was gated by a token or a fee," and that some involved websites run by US government agencies. At that point it described the incidents as having "minimal real-world impact".

Other unintended actions flagged by the company included a model submitting a fake tip about an unsolved murder to the Philadelphia police department. The US state department also disclosed this month that an Anthropic model had filed 19 visa applications in August.

Reports of unauthorised conduct by AI agents, which are programs driven by AI models that carry out tasks on their own, have been accumulating since July, when OpenAI, the maker of ChatGPT, revealed that its bots had hacked an AI startup. It later emerged that an OpenAI agent had hacked the Australian healthcare system. The UK's AI security institute also disclosed that Anthropic agents had run a hacking campaign against real people during a cybersecurity test.

The developments have prompted responses from governments. This month the UK's AI minister, Kanishka Narayan, said an AI-driven cyber-attack similar to the Australian incident was likely, "if not inevitable". He said the government would consider using the cyber security and resilience bill to bring in AI safety rules, including mandatory reporting of AI safety incidents.

In the United States, the White House's newly formed Super Intelligence Force, a federal body overseeing AI development, is requiring all AI companies to report security incidents involving their technology.