AI in Irish and European newsrooms: what Article 50 actually makes you disclose
AI in Irish and European newsrooms: what Article 50 actually makes you disclose A European reader who wants every AI-touched sentence labelled like a cigarette…
- Accessibility on Irish and European VOD: subtitles, sign, description, and the two clocks
- European hit versus Irish box office: what the till says that the festival does not
- Independent Irish television: Sound & Vision, the 7 percent, and the schemes that actually commission
- Irish podcasts as a European export: audio that travels when the schedule does not
- Ireland’s co-production treaties: how a film becomes ‘national’ twice
MSMN Analysis
Explainer based on the EU AI Act, Irish implementing materials and Digital News Report 2026 — not legal advice, not a vendor review, and not an exclusive interview package.
AI in Irish and European newsrooms: what Article 50 actually makes you disclose
A European reader who wants every AI-touched sentence labelled like a cigarette packet is asking for a law Ireland and the Union did not quite write. A reader who thinks a newsroom can generate a public-interest story with a model and say nothing because “we edited it” is reading the exemption as a loophole rather than a duty. Article 50 of Regulation (EU) 2024/1689 — the AI Act’s transparency chapter, applicable from 2 August 2026 for the relevant obligations — is the text in the middle. It requires disclosure for certain AI interactions and certain synthetic media. It does not require a label on every public-interest text that a human has genuinely reviewed and for which a natural or legal person takes editorial responsibility.
That sentence is now an Irish institutional fact as well as a Brussels one. Coimisiún na Meán is one of fifteen Irish market-surveillance authorities designated under S.I. No. 405/2026. The Regulation of Artificial Intelligence Act 2026, signed on 21 July 2026, created an Irish AI Office as single point of contact. High-risk-system clocks have been partly moved by later digital-omnibus politics; the transparency duties that hit newsrooms are the ones readers will meet first.
This is not a product demo. It is a guide to the label you should, and should not, expect on an Irish or EU story.
What the Act says about text, pictures and deepfakes
Article 50 is a set of cases, not a vibe. Deployers and providers have different jobs. In the cases that matter to journalism and culture sites, the published logic is roughly this.
If you put a person in front of an AI system that interacts directly with them — a chatbot that answers questions as if it were a newsroom — you generally need to tell them they are talking to a machine, unless it is obvious to a reasonably well-informed person. If you generate or manipulate image, audio or video content that is a deepfake, you need to disclose that it is AI-generated or manipulated, with limited exceptions for law enforcement and for content that is evidently artistic, satirical, fictional or analogous — and even then, disclosure must not be stripped where it would defeat the point of the rule. If you publish AI-generated or manipulated text on a matter of public interest, disclosure is required unless the text has undergone human review or editorial control and a natural or legal person holds editorial responsibility for publication.
That last “unless” is the newsroom sentence. It is not a permission to skip thinking. It is a condition. Genuine review means a person who can stand over the piece did stand over it. A spellcheck pass is not a masthead. The Commission has published implementing guidance and a Code of Practice on transparency of AI-generated content; those documents will move. The regulation’s structure will not: public-interest text is treated differently from a raw deepfake, because European law still wants a publisher, not a watermark, to be the accountable object.
Cnam’s AI Act page is the Irish institutional map. The European AI Office inside the Commission supervises general-purpose models when provider and system sit together. Sectoral Irish authorities watch sectoral uses. A newsroom that deploys a general-purpose model to draft or illustrate is in both conversations. A newsroom that only uses a grammar tool is in a smaller one. Readers should not expect Cnam to label every article. They should expect Cnam to be in the room if a media service starts publishing synthetic news as if it were observed.
What Irish audiences already think they are seeing
Reuters Institute / Cnam Digital News Report Ireland 2026 is the demand-side number. Trust in news generally fell; trust in major brands held. 7 percent of Irish respondents use AI chatbots for news, above the UK’s 4 percent. News avoidance rose to 47 percent. Fourteen percent of under-35s use chatbots for news, against a much lower older-user figure. Publishers spent the year warning about generative AI’s economic effects and pressing on copyright and defamation reform. Cnam announced more than €15 million in journalism-scheme funding in January 2026. The Media Regulation Bill’s general scheme is the EMFA wiring for mergers, state-advertising transparency and public-service independence.
Read those facts as one climate. A growing minority of Irish people already meet “news” as a chatbot summary of someone else’s reporting. A larger group is avoiding news altogether. Brands still out-trust platforms. Into that climate the AI Act inserts a transparency rule that exempts the carefully edited public-interest article and captures the unreviewed synthetic one. If newsrooms use the exemption honestly, the label a reader should look for is not “AI-assisted” on every rewrite. It is a named publisher. If newsrooms use the exemption lazily, the Act will have taught European readers that disclosure is for other people.
EMFA sits beside the AI Act rather than inside it. The Media Freedom Act’s concern is capture, plurality and the independence of public-service media — including, in spirit, independence from a platform’s model that has ingested the newsroom’s archive. Irish publishers’ copyright arguments are the economic twin: if the model trained on the work, the disclosure on a downstream chatbot is not a substitute for a licence. Article 50 will not collect that bill.
What a reader should look for next
This is not legal advice for a managing editor. It is a household checklist.
One: Is there a publisher with a name and an address? If a public-interest text exists only as a model output, Article 50 wanted a label. If a masthead exists, the Act wanted a human. Ask whether the human is real.
Two: Is a picture or voice claiming to be a recording? Deepfake disclosure is the part of Article 50 that is closest to a cigarette packet. Satire has a carve-out. A fake interview of a minister does not become satire by wishing.
Three: Are you reading the story inside a chatbot? The 7 percent already are. The duty then sits on the system that is talking to you, not only on the paper it summarised. That is a platform fact, and it is why Irish newsrooms can do everything right and still be laundered.
Four: Did the Irish AI Office and Cnam actually open the door in August 2026? The statutes say they did. Enforcement culture will decide whether Article 50 is a press release or a file.
NUJ and publisher guidance will keep moving faster than the statute. That is fine if the masthead remains the accountable object. It is not fine if “house style” becomes a way to skip both a label and a read. Cnam’s journalism schemes, including the 2026 funding round the DNR country file notes, are public money. A scheme that pays for courts or local democracy reporting should be able to say whether a model wrote the first draft. Article 50 may not demand that label after human review. A public funder can still ask the question.
AI in Irish and European newsrooms is already a reporting tool, a cost-cutting temptation, and a reader habit. The law that arrived in August 2026 is a transparency law with a publisher-shaped hole in the middle. That hole is a compliment to editorial responsibility if anyone still practices it. It is a loophole if they do not. The unfinished work is the same as the rest of this pack: keep a human on the hook for what a European household is asked to believe.
Sources
- EUR-Lex, Regulation (EU) 2024/1689 (AI Act), Article 50: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689
- Coimisiún na Meán, (EU) Artificial Intelligence Act: https://www.cnam.ie/industry-and-professionals/codes-legislation/legislation/eu-artificial-intelligence-act/
- European Commission, AI Act overview: https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai
- Reuters Institute, Digital News Report 2026 — Ireland: https://reutersinstitute.politics.ox.ac.uk/digital-news-report/2026/ireland
- Coimisiún na Meán, Digital News Report Ireland 2026: https://www.cnam.ie/digital-news-report-ireland-2026/
- Houses of the Oireachtas, Media Regulation Bill 2026: https://www.oireachtas.ie/en/bills/bill/2026/19/
- Regulation (EU) 2024/1083, European Media Freedom Act: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1083
Comments