Anthropic will watermark text made by its AI models like Claude

In short: Anthropic says it will add a built-in “watermark” to text created by its AI models, including Claude, to meet new European Union rules.

What happened

Anthropic says it will start watermarking text generated by its AI models, including Claude. A watermark here does not mean a visible stamp on the page. It means a hidden marker inside the writing that other systems can detect, like an invisible tag that says, “AI wrote this.”

The change is tied to the EU AI Act’s Transparency Code, which took effect on August 2. These rules tell AI companies to label content that was generated or edited by AI in a way that other software can recognize.

Anthropic says all models released after August 2 will automatically include watermarking for both text and files. For files, the company says it will use C2PA, which is a common method for attaching “content credentials” (think of it like a standard label that travels with a photo or document).

The company also says it will extend watermarking to older models. Anthropic says the marker is part of the text itself, so it should travel when you copy and paste the writing into another place, and it may survive some editing. Anthropic has not said how much editing it takes to remove the marker.

Why it matters

More people are running into AI-written content in school, at work, and online, and it is often hard to tell where it came from. Watermarks are one attempt to make that easier, especially for platforms, teachers, and publishers that want clearer labeling.

Source: TechCrunch AI