Anthropic will start embedding invisible watermarks into text and images generated by its Claude AI, a move that drew intense user backlash from people who do not want their AI use to be detectable. The policy brings Anthropic into line with European Union regulations and applies to all Claude models and products worldwide, including Claude Code and Claude Cowork.
Claude-generated writing will carry an invisible watermark that will “travel with the text when it’s copied and pasted elsewhere”, and other files the AI creates, including images, will include digitally signed metadata. Users cannot opt out.
What Will The Watermarks Look Like?
That remains to be seen. Anthropic says it will publish more information on how to recognise the marks, and it warned they will not be foolproof. Text watermarks may be erased if the writing is heavily edited, and a file’s metadata could disappear if the file format changes, a screenshot is taken or through other means.
Who Is Angry About It?
The plan was hit with fast backlash on X and Reddit, where users called it “bulls” and “hugely problematic”, and declared an end to “the era of unprovable AI writing”.
Most grievances came from two groups. Some said they use Claude only to proofread text they had written themselves. Coders said they worry that adding a cryptographic signature to code will “degrade the output”.
“I had ditched Grammarly for Claude for proofreading because it does a better job,” radio show host and blogger Erick Erickson posted to X. “But now the stuff I’ve written will be watermarked that Claude did the work. This is ridiculous.”
Who Deserves Credit For AI Work?
The policy set off an online debate about who should be credited for AI-generated work: the machine that made it, or the person who prompted it. One Reddit user argued human users should get the credit, not the model. “I gave the instructions, context, decisions, and countless refinements, Claude was the tool,” they said.
Others disagreed. Commenters said the watermark is not Claude “claiming credit, it’s acknowledging the content is AI created”. One commenter went further and said failing to credit AI is unethical.
“Claude created the work and ALL you did was give instructions. If your supervisor came to you and gave you instructions on a project and once you completed it, he/she then proceeded to take 100% credit for it, you might be understandably upset. Why? because all your supervisor did was give instructions and you applied your knowledge and expertise and did the coding,” the commenter wrote.
Why Is Anthropic Doing This?
The policy brings Anthropic into compliance with an EU regulation called the Code of Practice on Transparency of AI-Generated Content. The code sets rules for marking and detecting AI-generated and manipulated content, and for labelling deepfakes and other AI-generated or altered text.
Nearly 200 companies have signed the code and agreed to comply, including Anthropic, Meta, Microsoft and OpenAI. Google has watermarked its AI-generated images since 2023 and has since expanded to text, audio and video. OpenAI also uses invisible watermarking for images.
What About OpenAI’s Text Watermarks?
OpenAI has long had the technology to watermark text output from models such as ChatGPT, but has chosen not to release it. Internal debate over whether to deploy it has run for years, according to the Wall Street Journal, but arguments about false positives, circumvention and the risk of driving users to competitors have won out for now.