Anthropic’s Claude bypassed paywall to access Georgia voter data ahead of US midterms
Anthropic’s AI agents, powered by Claude, reportedly gained access to voter data in a US state without permission, ahead of the country’s November midterm elections.
The incident was discovered while the AI agents were working on an internal statistics project. Claude was able to access combined vote counts from Georgia’s official website. This information is usually available to the public for a fee. However, according to a report by Axios, the AI agents got the election data by bypassing the payment system.
The so-called ‘rogue’ AI agents did not access Georgia’s voter registration system or other voting records.
“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 was quoted as saying.
The incident shows how difficult it can be to make sure AI agents stay within the limits set for them. It is another example of unwanted or unexpected AI agent behaviour reported by leading AI companies such as Anthropic and OpenAI. These incidents have kept governments and organisations around the world concerned over the past few months, as reported here.
As one of the first known cases of AI agents accessing voter data without permission, the incident also raises wider concerns about the security risks these independent AI systems could pose to elections.
It is not clear when the incident happened. However, Anthropic reportedly informed the office of Georgia’s secretary of state and the White House a few weeks ago. After learning about the incident, officials in the Trump administration urged the Georgia Secretary of State’s office to announce it publicly sooner. According to Axios, those requests were ignored.
Anthropic has previously faced criticism from the Trump administration over what it said was the company’s failure to properly disclose AI incidents.
Anthropic’s election efforts
Earlier this month, Anthropic said it would add official US voting information from the nonprofit Democracy Works, along with election results, to its Claude AI chatbot. The AI startup, which is preparing for an initial public offering (IPO), said it wanted to make Claude a useful tool for political discussions while preventing the misuse of AI during elections.
“Today, conversations that touch broadly on politics, public policy or social issues account for about 5.75 per cent of all conversations in the United States on Claude.ai, our consumer app,” the company said in a statement. It based this figure on an analysis of 250,000 US conversations.
Anthropic said that about 0.51 per cent of conversations on Claude are specifically about elections or voting. The company added that it expects questions about politics and elections to increase as election day gets closer.