Anthropic will embed invisible watermarks in text and C2PA metadata in files generated by Claude AI. This move ensures compliance with transparency rules established by the EU AI Act.
The policy applies to all Claude models launched on or after August 2, 2026. Anthropic will implement these changes globally rather than limiting them to the European Union.
Watermarks will remain imperceptible to humans and persist through standard copying and pasting. Heavy editing may degrade the watermark's effectiveness.
Claude will attach signed C2PA metadata to image files including .png, .jpg, and .svg formats. This metadata indicates AI generation and helps detect subsequent tampering. Anthropic clarifies that these marks signal Claude processing but do not provide conclusive proof of AI authorship.