Anthropic's Claude AI Watermark: EU Compliance or Overkill?

Anthropic's new AI watermark for Claude text is drawing attention because it appears to go further than what European Union law requires. Experts say the EU's Article 50(2) does not mandate marking AI outputs for standard editing or minor changes to a user's words. However, Anthropic's watermark can appear in translations, summaries, and extensively edited text. One expert described it as using a "sledgehammer" instead of a "scalpel." Anthropic states it's implementing the watermark globally for EU AI Act compliance, as it lacks a region-specific application method. The company's approach has sparked backlash, with some users canceling subscriptions and developers creating tools to remove the marks. A key concern is that the watermark might imply Claude wrote the text, even when only used for editing. This is because the watermark works by guiding Claude's word choices to create a detectable pattern. This matters because it highlights a debate over how AI companies interpret and apply new regulations, impacting user experience and the future of AI transparency.