Anthropic published a blog post on Friday answering key questions about how it will watermark text generated by its chatbot Claude, following the company's earlier announcement that it would adopt the practice to comply with the EU AI Act's Transparency Code. The move has sparked debate among users, with some on Reddit calling it a conspiracy against innocent users and others defending it as a way to prevent deception. Business Insider reported that dozens of users on X have claimed to cancel their Claude subscriptions in response.

In the post, Anthropic explains that watermarking works by having Claude make "low-stakes choices"—such as picking between words like "overcast" and "grey"—to create a detectable pattern in its responses. The company says the watermark is "undetectable to the reader, but is detectable to anyone who has a key that encodes it," and that it does not affect output quality. Anthropic will use the SynthID-Text approach outlined by Google DeepMind in 2024 and plans to release a watermark detection API. It also distinguishes watermarking from AI detection tools like Pangram, which look for writing patterns, noting the two are fundamentally different.

On the question of evasion, Anthropic said light editing likely won't remove the watermark completely, while a full rewrite where every word is replaced will. The company added that if Claude only proofreads or lightly edits human-written text, there may be very little watermark to attach to. For code, the watermark will be minimal because the model must produce working code, though it may appear in arbitrary choices like comments. Anthropic also noted that other major model developers have signed the same Code of Practice and will implement their own watermarks.

Source: TechCrunch