NewsBite analysis · AI, Regulation · Software Release
Engadget
Anthropic is watermarking text generated by Claude to comply with EU law
Anthropic has announced a new watermarking feature for its Claude AI text generator to adhere to the European Union's AI transparency laws. This innovative method encodes patterns in the text that can only be decoded with a specific key, ensuring that users cannot easily discern the watermark visually. Unlike traditional methods, this watermarking does not degrade the quality or increase the cost of the generated text. However, it has limitations, such as the inability to determine the exact role Claude played in content creation if it only provided editing assistance.
Act nowConfidence: HighShort term horizon
Published 15 Aug 2026, 18:25 UTC3 min read
Visit original article ↗Lead image — Engadget
Summary
TL;DR
- Anthropic unveils watermarking for Claude AI text generation.
- Watermarking complies with upcoming EU AI transparency rules.
- Patterns in text are not visually distinguishable.
- Does not impact output quality or generation costs.
- API will allow users to verify AI involvement.
Interesting facts
- The watermarking method is inspired by Google DeepMind's SynthID-Text approach.
- Using the digits of pi as a key showcases an innovative technique in AI.
- Watermarked output includes text translations and edited materials.