Anthropic Found Hiding Tracking Code in Claude to Detect Chinese Users

Executive Briefing

• Reveals security researcher 'Thereallo' discovered hidden spyware-like code embedded in Claude Code's system prompt.
• Tracks user timezone and proxy usage to identify connections to Chinese AI labs suspected of model distillation.
• Justifies the covert tracker as an anti-abuse 'experiment' from March, claiming it was meant to be removed.
• Undermines Anthropic's ethical brand, which attracted users who valued its stance against mass surveillance.
• Warns that hiding classification signals inside AI tools with filesystem access erodes all future privacy claims.