Anthropic to Implement Invisible Watermarking for Claude Content
To align with European AI transparency mandates, Anthropic is rolling out machine-readable watermarks for text and images produced by its Claude models. While invisible to human users, these digital markers aim to provide platforms and individuals with a reliable method to identify content generated by artificial intelligence systems.
The company confirmed that future iterations of Claude—including the API, Claude Code, and Claude Cowork—will feature these identifiers from their launch date. For images, Anthropic has adopted the C2PA provenance standard, aligning its approach with industry leaders like Google, Adobe, and OpenAI. The strategy for text remains more opaque, involving an imperceptible watermark woven into the model's output that reportedly preserves the readability and original meaning of the response.
These technical adjustments arrive in response to the European Union’s AI Act, which became active on August 2nd. While the legislation provides a four-month grace period for products already on the market, Anthropic is treating the integration as a global standard for its entire ecosystem. Retrofitting existing models to support these tracking features remains an ongoing development process rather than an immediate deployment.
Comments (0)
No comments yet. Be the first!