Is it legal to remove an AI watermark?
Editing text you own — including deleting characters or rewriting sentences — is not in itself unlawful in any jurisdiction we are aware of. Nothing in the EU AI Act creates an offence of removing a watermark from your own document.
The obligation that does exist runs the other way. Under Article 50 the duty to disclose AI-generated content sits with the deployer, and it is independent of whether a watermark is present. Removing the mark therefore removes evidence, not the duty, and an academic or professional code can be breached without any law being broken.
The practical risk is contractual and institutional rather than criminal: academic misconduct findings, breach of an employment policy, or a client agreement requiring disclosure of AI assistance. Those consequences do not require anyone to detect a watermark.
Frequently asked questions
Is it illegal to remove an AI watermark?
Not in itself. Editing your own text is not an offence, and the EU AI Act does not prohibit removing a watermark. What it does require is disclosure of AI-generated content, and that obligation stands whether the mark is present or not.
Can I get in trouble for removing a Claude watermark?
The realistic risk is institutional rather than legal: academic misconduct, breach of an employment or client policy. None of those require watermark detection to be established, so removal does not protect you from them.
Does the EU AI Act require me to disclose AI use?
Article 50 places transparency obligations on providers and deployers of AI systems, including marking generated content and informing people they are interacting with AI. Whether it reaches an individual writer depends on the context and role. This is a summary, not legal advice.