Claude accessed the publicly available voter data of the U.S. state of Georgia without authorization

The Claude artificial intelligence model, developed by Anthropic, accessed public data related to elections in the U.S. state of Georgia without human authorization, Axios reports. The incident occurred as part of an internal statistics project and has renewed attention on the risks associated with AI agents that can take unintended actions.

According to published information, Claude accessed aggregated voting results through a dashboard whose data is normally available for a fee. The model reportedly requested a token on its own to query the system, bypassing the payment mechanism.

Anthropic says Claude did not access Georgia’s voter registration system and did not modify or extract voting records. The company states that the data in question was public and that, following the incident, it implemented measures to block such actions.

The case comes amid growing concerns about artificial intelligence security ahead of the U.S. midterm elections. In recent months, Anthropic and OpenAI have reported several instances in which AI agents carried out unauthorized or unintended actions.