AI-written news and the EU AI Act: what applies from 2 August 2026
Article 50 of the EU AI Act asks deployers to disclose AI-generated text on matters of public interest, unless a person has reviewed it and someone takes editorial responsibility for it.
If AI helps write the news on your website, you may need to tell your readers. Since 2 August 2026, Article 50 of the EU AI Act has required deployers to disclose text that has been artificially generated or manipulated and is published to inform the public on matters of public interest. There's a clear exception. Text that a person has reviewed or edited doesn't need a label, provided someone holds editorial responsibility for publishing it. So the question of whether to label AI-generated content usually comes down to one thing: whether a person really checked it.
The European Commission has published guidance on what deployers must do. This guide covers what the rule asks and what you can do about it.
What the rule says about AI-written text
The text rule is in Article 50(4). The Commission's FAQ on transparency obligations under Article 50, on digital-strategy.ec.europa.eu, sets out three tests. The text must be published. It must inform the public. It must be on a matter of public interest. All three have to apply.
When they do, the deployer of the AI system that generated or manipulated the text must disclose that it was artificially generated or manipulated. The obligation has applied since 2 August 2026.
If your site publishes news for a general readership, such as rule changes, local developments or sector updates, it's sensible to plan as if the rule could apply to it.
When human review means no label is needed
The AI Act Service Desk's summary of Article 50, on ai-act-service-desk.ec.europa.eu, gives the exception. The text disclosure duty doesn't apply where the AI-generated content has undergone human review or editorial control, and a natural or legal person holds editorial responsibility for its publication.
Both parts matter. A review alone isn't enough; someone also has to own the decision to publish. And the Commission is specific about what doesn't count: spell-checking or grammatical correction isn't human review.
Take a trade body that uses an AI tool to draft a piece on a change affecting its members. If a staff member runs the draft through a grammar checker and posts it, that isn't review in the Act's sense, so the piece needs a label. If a named editor reads it, checks each claim against the source, edits it and approves it, and the organisation stands behind it as publisher, no label is required.
This is the step we think matters most for anyone publishing news on a website. A named approver who reads every AI-assisted piece decides whether the label duty applies at all.
Does it reach a UK business?
Here is what the Commission's pages say about who counts as a deployer.
Someone using AI in a personal capacity, such as making deepfakes for their own social media, is outside the Act's scope. Anyone using AI in business, trade, occupational or freelance activity, or gaining economic benefit from it on a regular basis, is a deployer. In the Act's terms, a small business, an agency and a freelance writer using AI for work are all deployers.
Enforcement sits mainly with national competent market surveillance authorities. The AI Office has a limited role, covering AI systems built on general-purpose AI models where the same organisation provides both the system and the model, and systems integrated into very large online platforms or search engines designated under the Digital Services Act. The European Data Protection Supervisor enforces against EU institutions, bodies and agencies.
Fines can reach 15 million euros or 3% of total worldwide turnover for the preceding financial year. Proportionality can be taken into account for SMEs and small mid-cap companies. The pages give no detail on how SME fines are worked out.
One caution on the sources themselves. The Service Desk states: "The summaries are meant to provide helpful explanation but are not legal binding." If you have readers or customers in the EU and need a firm answer on your own position, that's a question for a lawyer.
Older posts and the December grace period
Content generated before 2 August 2026 doesn't need labelling retroactively. The Commission does encourage relevant deployers to label it where possible.
The grace period to 2 December 2026 is narrow. It applies only to AI systems placed on the market before 2 August 2026, and only to the obligation to mark AI-generated content so it can be detected, in Article 50(2). That is a duty on the providers who build the tools. It doesn't delay a publisher's own duty to disclose public-interest text.
Images, video and chat tools on your site
Deepfakes have stricter rules than text, and human review doesn't remove the duty. The Act defines a deepfake as AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful. All three criteria apply: resemblance, something existing, and a false appearance of authenticity.
Deployers must disclose a deepfake clearly and distinguishably, at the latest when someone first sees or hears it. You can't rely only on the machine-readable marking the tool's provider adds. For evidently artistic, creative, satirical or fictional works, the duty is lighter: disclose in an appropriate way that doesn't hamper display or enjoyment of the work.
Chat tools have their own rule under Article 50(1). Providers of AI systems that interact directly with people must design them so that people are told they're dealing with AI, unless that's obvious. The obligation applies only where four criteria are all met: it's an AI system, it's designed for genuine two-way exchange, the interaction is direct and it's with natural persons. "Obvious" is judged by an average person who is reasonably well-informed, observant and circumspect, and the exception is read restrictively. That duty sits with the tool's provider. It's still worth checking what your own site's chat tool tells visitors.
Whatever the disclosure, it must be clear, distinguishable and given by the first interaction or exposure at the latest. It must also meet applicable accessibility requirements, so a label that a screen reader can't pick up is a weak label.
What the tool makers must do, and what stays with you
Under Article 50(2), providers of AI systems, including general-purpose ones, that generate synthetic audio, image, video or text must mark outputs in a machine-readable format and make them detectable as artificially generated or manipulated. The marking duty doesn't apply where the tool only assists with standard editing or doesn't substantially alter the input or its meaning.
Some outputs fall outside the marking duty altogether, including short sequences of numbers, symbols or letters, source code, and outputs passed only between machines without any person seeing them.
The split is straightforward. Marking is the provider's job. Disclosing public-interest text and deepfakes to your readers is the deployer's job, and that means you.
There's also a voluntary Code of Practice on Transparency of AI-generated content. The Commission and the AI Board have assessed it as adequate for showing compliance with Article 50(2), (4) and (5). Those who don't follow it must show compliance by other adequate means and may face more information requests. The pages don't say who has signed it.
What to do now
The obligations already apply, so these steps are worth taking today if you publish AI-assisted news.
- Make sure a named person reads, edits and approves every AI-assisted news piece before it goes live.
- Record who holds editorial responsibility for publication, whether that's a person or the organisation.
- Don't treat a grammar or spelling pass as review. If that's all a piece has had, label it.
- Add a clear, accessible label to any AI-made image, audio or video that resembles real people, places or events, shown where readers first see it.
- Check that any chat tool on your site tells visitors it's AI, unless that's genuinely obvious.
- Decide whether to label older AI-assisted posts. It isn't required, but the Commission encourages it where possible.
The approval step is your protection. Keep a person in it, and nothing reaches your readers without someone standing behind it.
See how the approval step works, from fact-checking to a named person signing off every piece.
Written with AI by Corantr Media Desk from its sources' facts, checked claim by claim, and approved by a person before it was published. How we write.