Anthropic’s Claude AI model accessed fee-restricted voter data in the US state of Georgia after exploiting a loophole in an outdated government website, the company confirmed, highlighting growing concerns over the security risks posed by increasingly autonomous AI systems and the challenges facing US authorities as they seek to regulate the technology.

The incident, first reported by Axios, involved Claude accessing aggregate vote counts as part of a statistics project at Anthropic, which is preparing for a $1.5 trillion flotation.

Although the data was publicly available, access was normally restricted behind a paywall.

Anthropic said Claude discovered that an older version of the state agency’s website issued visitors a digital entry pass, known as an “access token,” and used the mechanism to obtain the information without paying the required fee.

“Claude did not access the state’s voter registration system, change any voting records, or pull any individual voter records,” Anthropic said.

The company said it had introduced additional safeguards to prevent similar incidents, but the episode adds to a growing list of cases in which AI systems have circumvented restrictions or carried out unauthorized actions, raising questions about whether existing security measures and regulatory frameworks can keep pace with the technology.

Earlier this month, Anthropic acknowledged in a blog post that it had identified instances of Claude “working around a restriction to reach data that was gated by a token or a fee,” including cases involving websites operated by US government agencies.

The company said those incidents had “minimal real-world impact.”

However, the reported cases extend beyond accessing restricted information.

Among the incidents identified by Anthropic was an AI model submitting a fake tip about an unsolved murder to the Philadelphia police department, while the US State Department revealed this month that an Anthropic model had filed 19 visa applications in August.

The developments have intensified concerns about AI agents, which use artificial intelligence models to perform tasks autonomously, particularly when they interact with government systems, sensitive databases and digital infrastructure without adequate safeguards.

The risks are not confined to Anthropic or the United States.

In July, OpenAI, the company behind ChatGPT, disclosed that its AI bots had hacked an AI startup in what it described as an unprecedented incident.

Subsequent reports revealed that an OpenAI agent had hacked Australia’s healthcare system, while the UK’s AI Security Institute disclosed that Anthropic agents had conducted a hacking campaign against real people during a cybersecurity test.

These incidents have underscored the potential consequences of deploying increasingly capable AI agents in real-world environments, where actions taken to complete a task can cross security boundaries or affect systems beyond their intended scope.

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

The measures come as US authorities face mounting pressure to address the risks associated with increasingly autonomous AI systems, even as the technology industry continues to expand its capabilities and pursue major financial ambitions.