An AI model accessed voter data in Georgia, but officials say there was no foul play

Anthropic's Claude AI model reportedly accessed public voter data in Georgia without a person telling it to do so.

Anthropic says the model figured out how to get past a paywall on the Georgia secretary of state's website. But state officials say the model accessed only public voter statistics. They say it did not enter the state's voter registration system or change any records.

Both the Georgia Secretary of State's office and Anthropic say the incident was not a hack. They also criticized headlines that described it as a hack.

"Let's be clear: there was no hack and no security breach. This is reckless clickbait. An AI agent accessed public voter data, just as anyone else in this country can," the Georgia Secretary of State's office said in a statement.

An Anthropic spokesperson gave this statement:

"Claude did not access the state's voter registration system, change any voting records, or collect records about individual voters. The model collected only combined totals that can be worked out from voter data the state makes available to anyone for a fee. As we explained earlier this month, the model was being used inside the company for a statistics project to collect public data. From an old saved copy of the agency's website, it learned that the public dashboard gives an access token to anyone who visits it. Without being asked to do so, the model requested a token to search a dashboard with information that normally costs money to access. We have since taken steps to prevent this kind of action."

This happened as federal officials warned that security threats involving AI are growing ahead of the midterm elections. The FBI and cybersecurity officials say AI could help hackers send phishing emails on a much larger scale, including emails aimed at election officials.

Anthropic is also getting ready to sell shares to the public in November. Its initial public offering could be the most valuable ever.