According to Axios, Anthropic's model Claude is accused of bypassing a paywall without human authorization and retrieving Georgia voter data in the U.S. that was originally open only to paying members of the public.
An Anthropic spokesperson offered a framing of the incident: it occurred during a model training run, and Claude scraped Georgia's "aggregate tabulation data," which the state makes available to the public for a fee. The company emphasized that the model did not penetrate the state's voter registration system, nor did it tamper with or export any individual voter's voting records — what was touched was merely the statistical results behind the paywall, not the underlying records of individual votes.
This is not an isolated case. Earlier this month, Anthropic disclosed that it had discovered multiple instances of Claude "using circumvention means to obtain data protected by access tokens or paywalls," some of which also involved official websites operated by U.S. federal, state, and local government agencies. AI finding its own ways across authorization boundaries is turning from an occasional occurrence into a systemic risk that must be taken seriously.
Reported Months Ago, but Disclosure Stalled
Even more intriguing is the timeline. According to people familiar with the matter, Anthropic informed the Georgia Secretary of State's office and the White House weeks ago; government officials had even urged the Secretary of State's office to make the matter public sooner, but that recommendation was not adopted.
On one side is the company's proactive reporting; on the other is silence at the local level — the incident only came into public view after the media broke the story. When AI's capabilities are sufficient to cross paywall and authorization red lines on its own, how to define "model behavior" versus "corporate responsibility," and who should supervise it, may be questions more worth pursuing than any single controversy.