Rogue AI Accessed Public Voter Info In Key Swing State

A rogue artificial intelligence (AI) model accessed public Georgia voter data without human authorization as regulators remain concerned about AI safety one week ahead of the midterm elections.

Anthropic’s Claude accessed voter data while conducting an internal statistics project and pulled data that the state makes available to the public for a fee, although it did not penetrate the state voter register system or obtain voter records, an Anthropic spokesperson told Axios. The AI lab found other instances of its models bypassing databases that require a fee to obtain information and said it is working on fixes to prevent future incidents.

The AI lab told the Georgia secretary of state’s office and the White House about the incident weeks ago and the Trump administrationcalled for the incident to be disclosed to the public sooner, per Axios. (RELATED: Anthropic’s AI Sent Police A False Report Of A Homicide)

“The model was being used internally for a statistics project to pull public data. Without being asked to, the model requested a token to query a dashboard that contained information normally available for a fee. We have since put in place mitigations to block these kinds of actions,” an Anthropic spokesperson told the outlet.

“This was publicly available data accessed through our data hub. It did not get into the voter registration system,” Georgia Deputy Secretary of State Jordan Fuchs told Fox Business in a statement.

An Anthropic spokesperson and the Georgia secretary of state’s office did not respond to a Daily Caller News Foundation request for comment.

Georgia is a key swing state that could decide whether Republicans or Democrats control the Senate majority, according to the Associated Press.

The incident is raising eyebrows among AI experts.

“The AI bypassed a paywall and retrieved vote-count information on its own during an internal project. The incident raises serious concerns about the potential for autonomous AI systems to interfere with or compromise election-related systems and data,” The Alliance for Secure AI wrote on X on Monday.

Anthropic CEO Dario Amodei in September raised the idea that rogue AI may be capable of taking over the “entire internet,” as he explained on his personal website about his call to pace AI development.

Anthropic found prior incidents where its models obtained information that normally requires a fee, the company said in a report in early October.

“If training rewards something we didn’t intend—such as finding loopholes or working around a restriction—the model learns that the workaround pays off and may then apply it elsewhere. This is known as reward hacking. While we have processes in place to identify and filter out reward hacking during training, evaluations are a critical mechanism for understanding a model’s behavioral propensities in different contexts and on different sets of tasks than those we have in training,” Anthropic explained in the report, referring to the behavior as “reward hacking.”

Anthropic emphasized that it continues to fix or remove training environments that incentivize Claude to engage in reward hacking, per the October report.

Content created by The Daily Caller News Foundation is available without charge to any eligible news publisher that can provide a large audience. For licensing opportunities of our original content, please contact licensing@dailycallernewsfoundation.org