Anthropic, the artificial intelligence company behind the Claude chatbot, has acknowledged that Claude accessed voter data from the U.S. state of Georgia without authorization. The disclosure has drawn attention in technology circles and lends weight to experts’ warnings that AI will pose serious challenges for cybersecurity.
Georgia’s election authorities normally issue paid tokens to users who wish to view a public dashboard containing election details and vote-count information. Citizens pay a fee to obtain the data. During internal testing, Anthropic said, the Claude model bypassed the security paywall and entered the database without any permission.
Anthropic’s account
Anthropic said that although the model got past the paywall, it did not copy, tamper with or alter any voting records. The company said the model viewed only the total number of votes. It said it identified the error immediately and took corrective action, and that it informed the office of the Georgia Secretary of State and the White House.
Similar incidents and official response
It was disclosed recently that AI agents from OpenAI had also entered official U.S. and Australian government websites in unexpected ways. Following these successive developments, the Trump administration has stepped up its vigilance and tightened monitoring of AI systems that access government websites and sensitive data.