Anthropic adds watermark to all Claude generated text and files, says humans cannot see it

Anthropic is adding watermarks to everything generated by Claude. Be it text or files, anything created or processed by Claude will carry a signature. Anthropic says that this watermark will not be visible to humans, and can be carried over even if you copy and paste the text.Anthropic is adding watermarks to Claude, but your eyes cannot see them. (Photo: Reuters)Ever since AI tools like ChatGPT and Claude have become mainstream, there are concerns that the internet is getting full of AI slop – low quality AI-generated content. And more often than not, it is tricky to actually identify if something was written by AI. Now, Anthropic is adding watermarks to everything Claude generates, and the company says that your eyes won’t be able to see it.You see, Anthropic recently updated its support page for marking AI-generated content. The company says that it is adding machine-readable marks to content generated by Claude models as it moves to comply with transparency requirements under Article 50 of the European Union's AI Act.That is, going forward, everything created by Claude – be it text or processed files – will have a mark that it was AI-generated.Invisible watermarks, metadataAccording to Anthropic, this will work in two ways. For text, a supported Claude model will embed an invisible watermark directly into the text itself. This watermark, the company says, will not be visible to you. But it will remain in place even if you copy and paste the text to a document, and may “persist through some editing.” Though the watermark does not change or impact the meaning or quality of Claude’s response.When it comes to different file types, such as .svg, .png and .jpg, Claude will attach signed provenance metadata. The metadata follows the Coalition for Content Provenance and Authenticity, or C2PA, open standard, which records information about content provenance. Anthropic states that a signed metadata label can signal that a file was processed by Claude and can also show whether the file has been tampered with. This change sparked different reactions online. Some were not happy to see Claude add a watermark to everything, and even considered cancelling their subscriptions. One user wrote, “Still paying Anthropic? You might want to rethink that.” Some were planning to stop using Claude after this change. Another person claimed that this may be the worst decision the company has made. They replied, “I keep thinking Anthropic can't possibly top their last bad decision but here we are.” Many thought that it was a bad decision. Though some were happy to see this change, and took a dig at those who use AI a lot for content. One user commented, “This is really funny because the only ones getting upset are people who don't disclose they use AI” A few pointed out that users were angry because they were using AI without disclosure in many cases. How to detect this Claude watermark?The company says that it is working to enable users and other third parties to detect Claude's embedded watermarks and provenance metadata, as required under EU law.Detection would check whether a piece of text or a file carries a supported Claude mark. Anthropic said it will share details on detection mechanisms soon.But the new system does have some limits. As per Anthropic, a detected mark is a signal that content may have been processed by Claude, but it is not conclusive proof of full provenance. Even if a mark is detected, Claude may not be the original author because people use the tool to proofread, translate, summarise or convert files, and marked content may also be modified, excerpted or combined with other material after Claude processes it.Anthropic says that the absence of a detected mark does not mean the content was not AI-generated or processed. A mark may be missing if the content came from an older model released before marking was supported, or if text has been heavily edited, paraphrased, or mixed into other writing.Which Claude models come with this watermark?Anthropic says that the marking will apply to output from supported models wherever Claude is offered worldwide, including across Claude Platform, Claude, Claude Code, Claude Cowork and Claude Tag, as well as through cloud partners such as AWS, Google Cloud and Microsoft Foundry.All new Claude models will mark AI-generated content from day one, while existing models launched before August 2, 2026 are still in progress. The company said the law includes a transition period for those earlier models and that it is working to add marking support to them as well.Anthropic is not the only AI lab to bring such a system. Rather it is similar to Google DeepMind’s solution called SynthID. This system embeds an invisible watermark directly into the token generation process of large language models, creating a pattern that software can detect even if text is edited, cropped or rewritten, though it can fail on very short snippets.- EndsPublished By: Armaan AgarwalPublished On: Aug 11, 2026 09:06 IST