Since 2 August 2026, Article 50 of the EU AI Act makes disclosure of AI-generated content a legal obligation for anyone publishing to EU readers [1]. The human-review carve-out in that same article turns a quality gate you probably already have into a compliance tool — but only if the review is substantive, not a rubber stamp [1].
What took effect on 2 August 2026 and who is in scope
Article 50's transparency obligations are now live, and they are not limited to high-risk AI systems [1]. Any AI system used in the four situations the Article covers triggers the rules, including ordinary blog workflows that reach EU audiences.
The Act applies globally to providers and deployers whose AI outputs are used within the EU, even without an EU entity or office [2]. A U.S. company whose content-generation tool reaches EU users is subject to Article 50 [3]. You do not need European infrastructure or a local subsidiary to fall under the regulation.
Noncompliance carries serious weight: fines of up to €15 million or 3% of worldwide annual turnover, whichever is higher [2]. The European Commission adopted guidelines on Article 50 on 20 July 2026, giving deployers less than two weeks to interpret the rules before they took effect [2].
The two rules that touch blog posts
Article 50(4) requires deployers publishing AI-generated text to disclose that fact when the purpose is informing the public on matters of public interest [1]. The trigger is the publisher's purpose, not merely the subject matter [1]. Most blogs qualify. If you publish explainers, industry analysis, product comparisons, or opinion pieces that touch on economics, health, technology, or policy, you are likely in scope.
Article 50(5) sets the standard for how disclosure must happen: given at the latest at first exposure [1]. Buried disclosures in footers or terms and conditions do not comply [4].
The human-review carve-out and why your approval step matters
The disclosure requirement does not apply where content has undergone human review or editorial control and a person holds editorial responsibility [1].
The review has to be real, not superficial or cursory approval [1]. A human who clicks "approve" without reading — the rubber-stamp autopilot flow — does not qualify. Meaningful sign-off does.
Off-to-Full autopilot keeps human approval before publishing. The reviewer who signs off is the person holding editorial responsibility. Set autopilot to Full without that substantive review, and you lose the carve-out. Keep a gate that requires reading, and you keep it.
What a compliant disclosure looks like
The EU AI Office published the voluntary Code of Practice on Transparency of AI-Generated Content on 31 July 2026 [3]. The Code provides three free EU icons, developed by the AI Office, distinguishing fully AI-generated content from partially AI-modified content [5]. These icons are free to use without attribution.
Use of the icons is optional, but the labelling requirement itself is not [5]. You may design your own label, but the Code's icons carry a degree of presumption of conformity for signatories [2]. Non-signatories face closer scrutiny. The Commission's final Code of Practice was published on 10 June 2026 and gives the practical route to compliance [6].
If you do not qualify for the human-review carve-out, adopt the EU icons rather than inventing your own.
The gap the rules do not close
Article 50 leaves several boundaries blurry. The deepfake definition covers only image, audio, or video that resembles existing persons, objects, places, entities, or events and would falsely appear authentic [1]. The draft Guidelines clarify that clearly fantastical content — dragons, humans flying unaided — falls outside [1]. Most blog cover images, which are neither photorealistic portraits nor obvious fantasy, sit in an unregulated middle space.
The assistive-editing exemption covers grammar correction and non-substantial alteration of input data [1][7]. Where AI moves from correcting your sentences to restructuring your argument, the exemption ends. The Code's taxonomy distinguishes "fully AI-generated" from "AI-assisted" content with different disclosure requirements, but interpretation is still settling [1].
Why this post is timed now, not driven by search volume
Exact-match search volume for "eu ai act transparency" is small. The bet is on fast-growing related queries: "ai act article 50", "ai-generated content labelling", "eu ai act disclosure". Law-firm guidance is still landing; most blogs are caught unprepared.
The data on current practice suggests why. Among AI-using marketers, 78% do not tell website visitors that AI helped create their content [8]. Only 22% disclose, up from 16% in 2025 [8]. Manual checking is more common — 89% of AI-using marketers manually check AI-generated content, up from 80% in 2025 [9] — but checking is not the same as holding editorial responsibility, and it is not the same as disclosing when disclosure is required.
Article 50 merges them.
Your practical checklist
Inventory which of your posts are AI-produced or AI-assisted. The Code's taxonomy distinguishes fully AI-generated from AI-assisted content, and your disclosure obligation differs [1].
Decide who on your team holds editorial responsibility. This person must be named, not assumed.
Set the approval gate so review is substantive before anything goes live. If you use Contentrails, match your autopilot level to the review capacity you actually have, not the capacity you wish you had.
Adopt the EU icons if you do not qualify for the human-review carve-out. They are free to use without attribution and carry a degree of presumption of conformity for signatories [2][5].
Please note: this is not legal advice. The guidelines are new, regulatory practice is still forming, and your specific situation may require counsel.
Additional resources
- EU AI Act — Transparency Obligations for Providers and Deployers of Certain AI Systems
- EU AI Act: Transparency Obligations Take Effect 2 August 2026 // Cooley
- EU Finalises Transparency Rules for AI-Generated Content | Paul, Weiss
Leave a comment: Is your current approval workflow substantive enough to qualify for the carve-out, or are you adding disclosure labels?
Sources
- The EU AI Act’s Transparency Rules: A Practical Guide to Article 50 (artificialintelligenceact.eu)
- cooley.com
- EU AI Act Article 50 Transparency Obligations Go Live (bakerbotts.com)
- EU Finalises Transparency Rules for AI-Generated Content (paulweiss.com)
- EU AI Act (orrick.com)
- technology.org
- artificialintelligenceact.eu
- ahrefs.com
- ppc.land
Originally published on the ContentRails blog.
Top comments (0)