Anthropic announced last week that it would embed invisible watermarks in AI-generated content, a move intended to bring the company into compliance with new EU rules on AI transparency. The watermarks were designed to tag Claude's output so it could be identified as machine-generated.
The safeguard met near-instant pushback. Within hours of the announcement, coders were touting overrides online, and some say they have already found workarounds to strip or obscure the markers. The speed of the response turned a governance feature into an arms race almost immediately.
The episode highlights a familiar tension in AI content provenance: any signal embedded in output can, in principle, be undone by a motivated user. For Anthropic, watermarks are central to how AI companies pitch themselves as responsible partners to European regulators, raising questions about whether such safeguards can hold up faster than the rules can be written.