The company confirmed that all models released after August 2 now feature automated watermarking capabilities. For file-based outputs, Anthropic has adopted the C2PA open standard, a move aimed at establishing a verifiable provenance chain. The system is designed to trigger across the entire Claude ecosystem, including the platform API, Claude Code, and Claude Cowork, ensuring consistent compliance regardless of the interface used.
In section Startups & Technology
Anthropic Embeds Invisible Watermarks Into AI-Generated Text
To align with the European Union's recent transparency mandates, Anthropic is embedding digital watermarks directly into text and files produced by its Claude models. This technical shift ensures that AI-originated content carries a persistent marker, surviving copy-paste operations and minor edits to satisfy the EU AI Act’s strict identification requirements.
While the company asserts that these markers persist through some level of manipulation, the precise threshold of editing required to strip the watermark remains undefined. This industry-wide pivot follows mounting pressure from regulators and public backlash regarding AI-generated misinformation. Other major players, including Google, Meta, and OpenAI, have similarly pledged to follow the EU’s transparency code as the legal landscape for generative tools shifts from experimentation to strict oversight.
Comments (0)
No comments yet. Be the first!