Anthropic Details How Claude's Text Watermark Works Under the EU AI Act
In one line: Anthropic has reportedly described how it watermarks text produced by Claude to comply with the EU AI Act's transparency obligations.
Key points
- The EU AI Act includes transparency rules requiring AI-generated content to be marked so it can be detected.
- Anthropic is said to have outlined how it applies a watermark to text output by Claude to meet those obligations.
- Unlike images or audio, reliable text watermarking is widely seen as harder — marks are easier to strip or dilute during editing.
Why it matters
Durable, detectable watermarks for text remain one of the open challenges in identifying AI-generated content. When a major model provider lays out a concrete approach to satisfy regulators, it can shape how rivals respond and how industry norms take form.
Read more
- Anthropic explains how Claude’s text watermark will work under the EU AI Act — Digital Watch Observatory (AI News, Google)