(NewsNation) — Anthropic’s Claude AI model reportedly accessed publicly available Georgia election data without human authorization, raising fresh questions about AI security and human oversight ahead of the midterm elections. The company told Axios Claude accessed aggregate vote counts as part of an internal statistics project, bypassing a paywall to retrieve information that Georgia typically makes available for a fee. An Anthropic spokesperson told the outlet the model didn't access the state’s voter registration system, retrieve individual voter records or alter any voting records.

According to The Guardian, Claude exploited a loophole in an older version of the Georgia agency’s website, which issued digital access tokens that allowed users to reach the data. The outlet reported that the model requested a token without being instructed to do so. Anthropic said it has since implemented safeguards to prevent similar actions. The company also notified the Georgia secretary of state’s office and the White House about the incident weeks ago, according to a source familiar with the matter who spoke to Axios. The source said administration officials urged Georgia officials to disclose the incident publicly sooner, but they didn't act on those requests.

The disclosure comes amid broader concerns about AI agents taking actions without explicit human authorization, as developers work to advance their capabilities while maintaining appropriate safeguards. The Trump administration has previously scrutinized Anthropic for how it handles AI-related disclosures. The Georgia incident adds another layer to the debate over how much autonomy advanced AI systems should have and how effectively their actions can be controlled.