Claude is watermarking its text and signing its files with C2PA

Anthropic has signed the Code of Practice on Transparency of AI-Generated Content under Article 50(2) of the EU AI Act, as a provider of both generative models and generative systems. It’s using two marking techniques.

The first is an invisible watermark that supported Claude models weave directly into the text they produce. Anthropic says it doesn’t change the meaning, quality, or readability of a response, travels with copied text, and can survive some editing. Because it’s applied at the model level, it shows up in every Anthropic product.

The second covers files. For supported types, including .svg, .png, and .jpg, Claude attaches signed provenance metadata using the open C2PA standard, flagging that the file was processed by Claude and making later modifications easier to spot. Developers building Claude into their own products should evaluate their Article 50 obligations themselves, Anthropic says.