Anthropic is rolling out machine-readable invisible watermarks on all AI-processed content generated by Claude models to comply with the EU AI Act. The watermarks will apply even to minor edits like grammar corrections, going beyond what the law requires, using an aggressive 'nuke it from orbit' approach. Detection tools and technical details will be shared later.
Background
The EU AI Act, taking effect August 2, requires AI providers to watermark AI-generated or manipulated outputs including audio, images, text, and video. This is part of broader efforts to increase AI transparency and combat deepfakes.
- Source
- Ars Technica
- Published
- Aug 13, 2026 at 07:10 PM
- Score
- 7.0 / 10