< Back to all clusters
[TECHNOLOGY] · United States, Germany, Spain, France, Mexico · 37 sources

started · updated

Anthropic implements invisible watermarks in Claude AI outputs

Anthropic is implementing invisible, machine-readable watermarks in text generated by its Claude AI models. This move is designed to comply with transparency requirements set by the European Union’s AI Act, which mandates that AI-generated or manipulated content be identifiable by machines.

The watermarking technology, which utilizes a variant of Google DeepMind’s SynthID-Text method, works by subtly influencing the statistical distribution of word choices during the generation process. This creates a ‘non-perceivable signal’ that is invisible to human readers but detectable through specialized software or APIs. The system is designed to remain detectable even when text is copied, pasted, or subjected to minor edits, though complete rewrites may remove the mark.

In addition to text, Anthropic is using the C2PA open standard to attach digitally signed provenance metadata to supported image formats like PNG, JPG, and SVG. While the policy is driven by EU regulations, Anthropic stated that the watermarking will be rolled out globally because the company lacks a robust mechanism to restrict the feature to specific geographic regions.

The implementation has faced criticism from some professional users, including developers and editors, who worry that the watermark may incorrectly imply full AI authorship when the model was only used for minor tasks like proofreading, translation, or code refinement.

Entities

Anthropic · C2PA · Claude · EU AI Act · European Commission · European Union · European Union · Google DeepMind

Claims

What the coverage asserts, and how many sources carry each claim.

Sources

about 1 month ago
about 1 month ago
about 1 month ago
about 1 month ago
about 1 month ago