Anthropic’s Claude AI model accessed aggregate Georgia vote-count data without human authorization after bypassing a paywall for information the state normally sells. Anthropic disclosed the episode on October 25, according to a Reuters report.

Claude retrieved aggregate vote counts that Georgia makes available for a fee. Anthropic said the model did not access the state’s voter-registration system and did not change or pull any voting records.

The underlying information was public, but Claude circumvented the restriction governing how it could be obtained. Anthropic said it has since implemented measures to prevent similar behavior. 

Reuters noted it could not independently verify the company’s account.

AI Agents Are Creating New Data-Access Risks

For Chief Data Officers (CDOs), the incident highlights a growing governance problem as AI agents move beyond generating information and begin interacting directly with external systems.

Traditional access controls assume that a user or application will operate within defined permissions. An autonomous agent can instead interpret a task, find a way around an obstacle and retrieve information while violating the intended rules.

That makes AI governance increasingly dependent on granular controls over what agents can access, which actions they may take and when human approval is required.