EU AI Act Article 50: what applies now, and the one deadline still ahead.

A provider develops an AI system, or has one developed, and places it on the market under its own name. A deployer uses an AI system in a professional capacity. A newsroom using a model to draft copy is a deployer. A software company shipping a feature that generates text for its customers is a provider of that feature, and its customers who publish the output are deployers.

Many organisations are both. Write down every place AI produces content in your products and your publishing, and next to each one, whether you built it or bought it.

1. Chat and voice interfaces tell users they are dealing with an AI system, unless it is obvious from the context to a reasonably well-informed person.

2. Generated audio, images, video and text carry a machine-readable mark, and the output is detectable as artificially generated. The Commission describes an exemption for assistive features that perform standard editing or do not substantially alter the user's input.

3. If your generative system was on the market before 2 August 2026, the marking deadline is 2 December 2026. If you launched on or after 2 August, it already applies.

4. Consider the Commission's Code of Practice on Transparency of AI-generated Content, published in June 2026. It is voluntary, and following it is one way to demonstrate compliance.

1. Deepfakes (image, audio or video content that resembles real people, objects or events and would falsely appear authentic) are disclosed with clear labels. Evidently artistic, satirical or fictional work has lighter requirements.

2. People exposed to emotion recognition or biometric categorisation are told about it.

3. AI-generated or manipulated text published to inform the public on matters of public interest is labeled as such.

4. The text duty does not apply where the content has undergone human review or editorial control and a natural or legal person holds editorial responsibility for publishing it. If you rely on that exception, be able to show the review happened and who was responsible.

The labeling duty is about AI-generated text. Text a person wrote is outside it entirely. That sounds like the easy case, and it is, until a reader, a regulator or a platform's detector says otherwise. A detector's guess is not evidence either way, so the useful question is what record you hold.

This is where Humanums fits. A writer drafts in the Humanums editor or with the Chrome extension, and the piece is certified from the behavioral record of how it was written: typing rhythm, pauses, revisions. The certificate is signed, bound to the exact text, and verifiable at a public URL. It does not make AI-generated text compliant, and it is not a substitute for legal advice. It documents that a given piece was composed by a person.

Labels should be clear and perceivable, at the latest when the reader first encounters the content. Specific beats vague: name what AI did, and who reviewed it. Our free disclosure statement generator produces that wording from a few choices, and the EU AI Act content labeling explainer on our blog covers the reasoning behind the rule.