Dario Amodei’s Anthropic is adding watermarks to AI-generated content in order to comply with a strict new transparency law in Europe — drawing gripes from some users.
The AI giant, known for its Claude chatbot, linked the move to the European Union AI Act’s transparency code — aimed at helping people distinguish human-made content from the AI-generated variety — which took effect on Aug. 2.
The watermarks appear automatically in AI-generated text, images and files, according to a support page on Anthropic’s website.
“When a supported Claude model generates text, it weaves an imperceptible watermark directly into the text itself,” the company’s website said. “You won’t see it, and it doesn’t change the meaning, quality, or readability of Claude’s response.”
Anthropic added that the watermarks will “travel with the text when it’s copied and pasted elsewhere, and may persist through some editing.”
The watermarks will be implemented “wherever Claude is offered, worldwide,” including within the Claude Platform API, Claude Code, Claude Cowork and Claude Tag.
Anthropic acknowledged that its detection program has “limitations” and may not always identify AI-generated content, especially in cases where text has been “heavily edited, paraphrased, translated, or mixed into other writing” or the “passage is very short.”
The European Commission, the EU’s digital watchdog, has said it had “strong backing” from the tech industry in its push to clearly label AI-generated content so that users aren’t misled.
As of July 31, the EU had secured commitments from nearly 200 companies to comply with its transparency rules.
Aside from Anthropic, AI giants like Sam Altman’s OpenAI, Mark Zuckerberg’s Meta, Google and Microsoft all signed on to the pledge.
“By signing the code, providers and deployers of generative AI systems signal their intention to promote public trust in AI and to mitigate deception and misinformation,” the EU said in a press release.
The efforts by Anthropic and other companies come amid heightened concerns about the rise of so-called “AI slop” and AI-generated content being passed off as authentic.
Last month, a debut novelist’s $2 million book deal was canceled after a publisher became concerned that he had used AI to write the manuscript.
Still, Anthropic’s new policy irked some commenters, with one calling it “total bulls–t.”
Others pointed out that tech-savvy consumers may be able to find workarounds.
It’s “hard to see what Claude’s watermarking approach really solves,” one person wrote on X.
“There are already plenty of free and open-source models without watermarks, so anyone who really wants to avoid them can simply use something else.”
