Anthropic's Hidden Code Sparks Trust Crisis

Anthropic quickly removed a tracker that was secretly monitoring Chinese users in its Claude Code service after a security researcher exposed the hidden functionality. The AI firm acknowledged the code, embedded using "prompt steganography," was an experiment initiated in March to prevent unauthorized resellers and counter distillation attacks.

Despite Anthropic's explanation that the code was meant to protect against intellectual property theft and account abuse, privacy advocates and users expressed alarm. The incident has eroded trust, with the developer who flagged the issue noting Anthropic could have opted for transparent methods instead of covert tracking.

The backlash has been swift, with Alibaba banning its employees from using Claude Code due to security risks. This action highlights the growing tensions and competitive pressures between US and Chinese AI firms, particularly concerning model distillation, which Anthropic is advocating to be deemed illegal.

Anthropic is urging the US government to implement stricter export control policies to prevent Chinese entities from gaining a technological edge through distillation. The company argues that such attacks pose a threat to national security and industry safety standards, emphasizing the need for collaborative solutions.