Forced Compliance: Anthropic Embeds Invisible Watermarking in Claude Texts

European regulation is forcing developers to change their underlying architecture. On August 13, 2026, it was revealed that Anthropic had begun embedding an invisible, machine-readable code (watermarking) into the text responses of the Claude model.

This is a direct technical response to the transparency requirements (Transparency Code) from the EU AI Act that have come into force. Commercial users who used AI for covert report generation, spam, and B2B copywriting are openly expressing their dissatisfaction, as their content is now easily trackable and classifiable. However, for the lab, a fine of 3% of its turnover is more terrifying than losing a portion of its shadow traffic. The era of anonymous generative content is definitively over. Companies without their own local (on-premise) open-weight models will no longer be able to pass off synthetic text as human labor without legal risks.

Source: Anthropic / European Union / TechCrunch