Claude will introduce a secret watermark in all its texts: goodbye to using AI without other people knowing

Anthropic is preparing a change for Claude that, while not seeming all that major, will change AI forever. Because of its commitments to the Code of Practice under the EU AI Act, the company will soon require its supported models to include an imperceptible signal in the text they generate. The watermark will stay with anything we copy and paste, and it may even survive edits. This transparency measure comes as the European Union is tightening obligations around AI.

Claude Download

An invisible mark that will stay with our text

According to Anthropic, the system will operate at the model level and, therefore, will be present in Claude, Claude Code, Claude Cowork, the API, and Claude Tag. The company will roll out watermarking worldwide, even though the measure originates in Europe.

Anthropic says the mark will be imperceptible and won’t interfere with the meaning, quality, or readability of the text it generates. It also explains that detection mechanisms, including third-party ones, will be released later in the technical documentation needed to detect the hidden patterns.

While OpenAI works on systems to detect AI-generated images and more and more services label posts and content as “created by AI,” the trend is clear. We don’t want to read content that doesn’t have someone behind it we can trust. There is still a long, long way to go before we can confidently say that a piece of content was generated by AI without false positives, if that’s even possible.