If you only read headlines from earlier today about Claude’s access to Georgia voter data, you might have gotten the impression that Anthropic is revealing something super scary — or even some sort of threat to democracy. I don’t necessarily claim otherwise, but personally this incident doesn’t exactly send a chill down my spine. In fact, it almost makes me feel for Claude.
And that’s because essentially all the model did was break a paywall.
Axios, which broke the story, called it “a release that is sure to add to growing concerns about AI security ahead of the midterm elections.”
However, in a Fox Business story — because of course Fox Business picked this up — Jordan Fuchs, Georgia’s deputy secretary of state, was quick to clarify that “This was publicly available data accessed through our data hub. It didn’t get into the voter registration system.”
Anthropic told Fox Business that its model came in “aggregated numbers that can be derived from voter data that the state makes available to everyone.”
Per Axios, this was all part of an internal statistics project at Anthropic. This instance of Claude autonomously “requested a token” to “request a dashboard” intended to access the paywalled information.
The token in question was not an AI token, but probably a JSON web token or something similar – a behind-the-scenes bit of code that allows or denies access to data on websites. Sounds like Claude has figured out how to easily hack the paywalled system for the secret coin vending machine, if you want, and it worked. For his trouble, this Claude agent got some publicly aggregated election data, which usually costs money.
Axios says Anthropic told the Georgia Secretary of State’s office, and for good measure the White House about this incident. Axios’ anonymous source for the story says Anthropic ignored a request from the White House for disclosure, which apparently led to that source informing Axios.
Judging by the fact that there are hundreds of millions of saved links on Archive.is, sidestepping paywalls for free access to information would seem to be a fairly popular violation of the law, and for the record, the FBI tried to crack it. But Claude did not exactly hack the voting machine with this security incident. In fact, this was probably the most relatable AI security incident to date.