Anthropic Implements AI Watermarks to Comply With EU Law

Anthropic implemented an imperceptible watermark on text generated by its Claude AI assistant to comply with the European Union's AI Act. The watermark is designed to distinguish AI-generated material from human writing and remains present even after the content is copied, pasted, or edited.

The move triggered a wave of subscription cancellations, particularly among developers and consultants. These professional users expressed concern that watermarks appearing in client deliverables, such as code or documentation, could raise questions about authorship or trigger contract penalties, even in cases where the AI was used only for translation or proofreading.

While some industry observers argue that watermarking increases transparency and prevents model degradation, affected users have begun migrating to competitors including Cursor, Grok, and various Chinese AI providers. Anthropic maintains that the watermark does not affect output quality and denies that there is a broader trend of increased cancellations.