Anthropic to Watermark Claude's Texts with SynthID-Text
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisis ton rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
Anthropic plans to add invisible watermarks to texts produced by Claude and has explained its approach. The publisher intends to use a variant of SynthID‑Text, a technology from Google DeepMind, and to integrate this text marking with C2PA support for images. The goal is to comply with European transparency rules that require machine-readable marks on synthetic content.
The EU mandates machine-readable markings, Anthropic aligns
The European AI law requires that artificially generated audio, images, videos, and text include machine-readable marks. To meet these requirements, Anthropic is introducing a watermarking feature and is also leveraging the existing C2PA support for images processed by Claude.
Claude will transition to an invisible watermark based on SynthID‑Text
Anthropic intends to integrate invisible watermarks into the texts produced by Claude to comply with European transparency requirements. On Friday, the company stated that the text marking method used would be an adaptation of the SynthID‑Text technique. This technology, developed by Google DeepMind, is open-source and allows for the generation of detectable patterns by manipulating word probabilities. This process is designed to identify synthetic texts without making the watermark visible to the reader.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.