Blog

The EU AI Act's Article 50: what changes for people who publish text

Since 2 August 2026, AI-generated content must carry machine-readable marks and some AI-written text must be disclosed. What that means in practice, and how to document human review.

Article 50 of the EU Artificial Intelligence Act sets transparency rules that apply from 2 August 2026. They affect two groups: the companies that provide AI systems, and the people and organisations who publish what those systems produce. This is a plain-language summary, not legal advice.

Providers must mark AI output

Providers of systems that generate text, images, audio or video must mark the output in a machine-readable way, so it can be detected as artificially generated. In practice this means watermarks and metadata such as C2PA Content Credentials. Systems that only assist with standard editing, without substantially changing the input, are exempt.

Publishers must disclose some AI text

If you publish AI-generated or manipulated text to inform the public on matters of public interest, you must disclose that it was artificially generated, unless a person reviewed or edited it and someone holds editorial responsibility for the publication.

Three practical consequences

  1. Don't strip the marks. Watermarks and metadata are how the system is meant to work. Probator shows hidden characters and AI marks but never removes them.
  2. Marks get lost anyway. Copy-paste, format conversion and paraphrasing remove many marks. Detection remains necessary for text that arrives unmarked.
  3. Document your human review. If you rely on editorial review, keep a record of it. A Probator certificate can record a first check of a draft, a later check of the final text, who reviewed it, and how much of it changed.

How Probator helps

Whether a disclosure is required depends on who you are and how you publish. Check with your adviser.

Check your own text