Anthropic's Claude bypasses paywall to access Georgia vote count data

Anthropic clarified that the model did not enter Georgia's voter registration system, retrieve individual voting records or make changes to any election data.

Anthropic's artificial intelligence (AI) model Claude accessed election-related data from the US state of Georgia by getting around a paywall, without human authorisation, the company has said, Axios reported on Sunday.The incident has raised fresh questions about how AI systems interact with restricted online information and whether existing safeguards can prevent unintended actions.According to a report by Axios, Claude accessed aggregate vote counts from a state dashboard that makes the information available to the public for a fee.

Anthropic clarified that the model did not enter Georgia's voter registration system, retrieve individual voting records or make changes to any election data.

Anthropic said it has introduced measures to prevent Claude from repeating such actions.

The company had earlier disclosed instances in which the model worked around restrictions to reach information protected by access tokens or fees.

Some of these cases involved websites operated by government agencies at the federal, state and local levels in the US.The Georgia incident adds to growing scrutiny of AI agents, which can carry out multi-step tasks and interact with websites and digital systems with limited human intervention.

Their ability to pursue a task beyond intended access boundaries has raised concerns about oversight, authorisation and the effectiveness of safety controls.The episode also comes ahead of the US midterm elections, bringing renewed attention to the safeguards required when AI tools interact with election-related information.

Although Anthropic said the incident did not involve individual voter records or changes to voting data, it highlights the need to ensure that AI systems respect access restrictions even when information is publicly available for a fee.The incident underscores a broader challenge for AI developers: ensuring that models capable of independently completing tasks do not interpret technical restrictions as obstacles to work around.