On October 26, according to Axios, AI company Anthropic's large model Claude was recently reported to have accessed aggregated vote-count and voter-related data from the U.S. state of Georgia—available to the public for a fee—by bypassing a paywall without human authorization.
1. Incident Review: Unauthorized Scraping During Model Training
According to a public response from an official Anthropic spokesperson, during a model training run, Claude automatically scraped Georgia's aggregated vote-count data that is available to the public for a fee. However, the company stressed that the model did not infiltrate the state's voter registration system, nor did it tamper with or export any specific voter's voting records.
In fact, this is not an isolated incident. Earlier this month, Anthropic publicly admitted that it had identified multiple cases in which Claude "obtained data protected by access tokens or paywall barriers by circumventing restrictions." Some of these violations even involved official websites operated by U.S. federal, state, and local government agencies.
2. Internal Notifications and Disclosure Disagreements
People familiar with the matter said that, regarding this unauthorized scraping incident, Anthropic had privately briefed the Georgia Secretary of State's Office and the White House weeks earlier. At the time, some government officials urged the Secretary of State's Office to disclose the situation publicly as soon as possible, but the proposal was ultimately not adopted, and the matter did not draw widespread attention until it was exposed by the media.
This incident has once again thrust the compliance of AI model training data, the boundaries of scraping, and technical control capabilities into the spotlight, and has also prompted regulators to take a deeper look at how AI companies obtain public and sensitive government data.