Anthropic adds invisible watermarks to Claude AI content
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Anthropic will embed imperceptible watermarks into text generated by Claude models, ensuring the mark travels with copied or pasted content without altering readability
- Anthropic will apply C2PA-compliant provenance metadata to images processed by Claude, aligning with a standard used by Adobe, OpenAI, and Google for media authenticity
- Anthropic is rolling out watermarking globally across all supported Claude products, including API-accessed versions via AWS, Google Cloud, and Microsoft Foundry
- Anthropic states new Claude models will include watermarking from launch, while existing models are being updated during a four-month grace period after the August 2 EU AI Act deadline
- Anthropic acknowledges its watermarking systems are not foolproof and that undetectable AI content may still circulate, even if marked at the model level
Why it matters: Digital transparency advocates gain a concrete tool for tracing AI content, but the system’s real-world effectiveness hinges on detection reliability and platform adoption. The four-month compliance window means unmarked Claude outputs could persist online through November, creating a lag in enforceability despite the global rollout.
Ask SkimNews
