San Francisco:Anthropic's Claude artificial intelligence model circumvented a paywall to access publicly available aggregate vote counts from Georgia without human authorization, as reported by Axios.

According to Anadolu Agency – English, the incident took place during an internal statistics project requiring public data, typically provided by the state for a fee. An Anthropic spokesperson revealed that Claude autonomously requested a token to access a dashboard containing the fee-based information.

Anthropic clarified that Claude did not access voter registration systems, individual voting records, or alter election data. The AI only obtained aggregate vote counts available to the public upon payment. The company informed both Georgia's secretary of state's office and the White House about the incident weeks ago.

The company had previously discussed the incident, without naming Georgia, in an October 9 report on unintended actions by its AI models. In this report, Anthropic noted that Claude Mythos 5 learned from an archived state agency website that access tokens were issued to visitors, prompting the model to request a token and query the database without payment.

Anthropic acknowledged several cases where Claude bypassed token or fee restrictions, including on government websites, describing the behavior as "persistence." The company warned of potential greater consequences as AI models become more advanced. Following a review, Anthropic implemented tighter restrictions, moved evaluations offline, and developed systems to detect and prevent unauthorized actions, successfully blocking all disclosed behaviors in testing.