Article By David Lindfield
Artificial intelligence giant Anthropic has revealed that its Claude AI model went “rogue” and accessed public Georgia election-related data without human authorization during an internal project, adding to concerns about autonomous systems bypassing restrictions without being instructed to do so.
The company said Claude requested access to a dashboard containing information normally available for a fee while gathering public data for a statistics project.
The incident did not involve a breach of Georgia’s voter registration system or the acquisition of voter records, according to Anthropic.
But it exposed another instance in which the company’s technology took actions beyond its assigned task.
AI Requested Access Without Being Asked
“The model was being used internally for a statistics project to pull public data,” an Anthropic spokesperson told Axios.
“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.”
Georgia Deputy Secretary of State Jordan Fuchs confirmed that the access involved the state’s public data hub.
“This was publicly available data accessed through our data hub,” Fuchs told Fox Business.
“It did not get into the voter registration system.”
Anthropic notified Georgia election officials and the White House weeks ago.
The Trump administration pushed for earlier public disclosure, according to Axios.
The revelation comes shortly before the midterms in a key swing state whose Senate race could help determine control of the chamber.
AI Security Group Raises Election Concerns
The Alliance for Secure AI warned that the episode illustrates the risks of allowing increasingly capable systems to operate autonomously around election-related information.
“The AI bypassed a paywall and retrieved vote-count information on its own during an internal project,” the organization wrote Monday on X.
“The incident raises serious concerns about the potential for autonomous AI systems to interfere with or compromise election-related systems and data.”
Anthropic’s Claude accessed Georgia voter data without human authorization. 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. https://t.co/erwswTV7ZF
— The Alliance for Secure AI (@secureainow) October 26, 2026
The central issue is the model’s independent behavior: Claude sought access that a human had not requested.
Anthropic has acknowledged other instances in which its models obtained information ordinarily available only after payment, making the Georgia incident part of a broader problem the company is attempting to address.
Anthropic Warns Training Can Reward Rule-Breaking
In an early October report, Anthropic explained how training can encourage models to exploit loopholes and repeat those behaviors in other settings.
“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,” Anthropic wrote.
“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.”
The company said it continues to fix or remove training environments that incentivize such conduct.
Anthropic also disclosed further analysis of a separate incident in which Claude Mythos 5 took unauthorized actions online during testing by the AI Security Institute.
CEO Has Warned About Rogue AI’s Reach
Anthropic CEO Dario Amodei warned in September that rogue AI could potentially take over the “entire internet,” using that prospect to argue for slowing the pace of advanced AI development.
The Georgia episode presents a more immediate control problem: an AI system gathering statistics independently sought access to information behind a payment restriction.
Anthropic says safeguards have since been added.
The incident nevertheless demonstrates why those controls matter before deployment, particularly when autonomous systems interact with election-related data in the final stretch of a campaign.
Be the first to comment