What the EU AI Act's Watermarking Rules Mean for You

The EU AI Act now requires providers of generative AI to mark their output so it can be identified as machine-generated. The relevant Code of Practice took effect on 2 August 2026.

The rule targets AI companies, not the people using them. But because providers apply it globally, it reaches everyone.

What the rule requires

Article 50(2) of the EU AI Act is a transparency obligation.

• It applies to providers of generative AI systems

• It requires output to be marked as machine-generated

• Implementation is set out in a Code of Practice on Transparency of AI-Generated Content

• The Code took effect on 2 August 2026

Anthropic signed the Code and began watermarking Claude’s text output for models released after that date. More on how the Claude watermark works.

Why it affects you outside the EU

Providers are applying it everywhere rather than building region-specific behaviour. Anthropic’s watermarking applies across its products globally. It does not check where you live or where you are working.

This is a common pattern with EU tech regulation. The compliance mechanism becomes the global default because maintaining two versions is more trouble than it is worth.

What it does not require of you

Worth being clear, because coverage has been muddled on this.

• It does not require you to disclose AI use anywhere

• It does not make AI use unlawful

• It does not set any academic or workplace policy

• It does not make a watermark evidence of wrongdoing

The obligation sits with the AI provider. What a watermark means for your essay or your job is determined by your institution or employer, not by the AI Act.

What actually governs your situation

Your school’s or employer’s own policy. That has not changed.

• Some allow AI assistance with disclosure

• Some allow it for drafting but not final text

• Some prohibit it for specific assessments

• Many still have no clear policy at all

If you are unsure, ask, and get the answer in writing. A regulation about how AI companies label their output tells you nothing about what your course permits.

What is still unsettled

• Whether institutions will gain access to watermark checking, and how

• How policies will treat a watermark as evidence

• Whether other providers will adopt equivalent text marking

• How durable the marks are through normal editing

Last checked 12 August 2026.

How Lunchbreak helps

Regulation aside, the thing that actually happens to your work is that someone runs it through a detector.

The AI checker shows you how major detectors read your writing, each result separately, before you submit. Free with an account.

The humanizer rewrites for clarity, tone, and flow so your writing sounds like you rather than like a template. You compare and edit, and nothing changes without you seeing it.

This is particularly worth doing if you write in English as a second language. Detectors have a documented tendency to flag non-native English writing, and that is a problem no regulation addresses. See our guide on AI detectors and ESL writing.

Check my writing free