Summary
From 2 August 2026, a life-sciences desk that publishes with a generative system has five outputs to sort: a disease-education page, a product page, a press release, a site chatbot, and a LinkedIn post or a newsletter. For each one the question is concrete. Does Article 50(4), second subparagraph, of Regulation (EU) 2024/1689 require a disclosure a person can perceive, and what record of review would let the team skip it? [1]
The statute puts the duty and the skip together. Deployers of an AI system that generates or manipulates text published to inform the public on matters of public interest shall disclose that the text has been artificially generated or manipulated. "This obligation shall not apply where the use is authorised by law to detect, prevent, investigate or prosecute criminal offences or where the AI-generated content has undergone a process of human review or editorial control and where a natural or legal person holds editorial responsibility for the publication of the content." [1] The two review conditions are cumulative. The criminal-offence authorisation is the other exit.
The Commission guidelines of 20 July 2026 already split the two pages this desk publishes most often. Text that must be labelled unless human review or editorial control applies includes "AI-manipulated parts of a lifestyle-website article comparing the effects of various diets on a particular disease in middle-aged women." Text falling outside Article 50(4) includes "AI-manipulated text that is part of a company's advertisement or product descriptions (not including any claims related to e.g. health, consumer safety or sustainability)." [2]
Point 131 uses three elements. The text is published. Its purpose is to inform the public. The subject is a matter of public interest. Published means an indeterminate, fairly large number of unrelated readers can open it, at once or in succession, paid or unpaid. Subscriptions are the illustration of payment. A closed private group that is too small or insignificant, private correspondence, and an internal corporate network stay unpublished. Informing means communicating knowledge, opinions or facts, so a one-line navigation label sits outside the element. Public interest means a matter relevant to society that merits debate or scrutiny. The list names public health, environmental protection and consumer safety, with politics, administration, justice, rights, security and debated economic or scientific developments. What counts can change. [2]
The parenthetical is part of the product-description example. Claims related to health, consumer safety or sustainability stand outside the illustration of text that falls outside the duty, and the example sets no rule that every such claim is public-interest text. [2]
Two further sentences do later work. The in-scope list includes "AI-manipulated corporate reports published on a listed company's website containing investor information." The out-of-scope list includes "News summary generated by a chatbot that is only available to the user that prompted the chatbot." [2] A press release, a public transcript and a LinkedIn post go unnamed.
Article 50 applies from 2 August 2026. Article 111(4), inserted by Regulation (EU) 2026/1744, requires providers of AI systems, including general-purpose AI systems, that generate synthetic audio, image, video or text and that were placed on the market before 2 August 2026, to take the necessary steps to comply with Article 50(2) by 2 December 2026. [3] December belongs to that provider mark. The text label keeps August. The FAQ of 24 July 2026 repeats public health on the point 131 list. [4]
This reading is informational, not legal advice.
Four duties, and the date that belongs to one of them
Four sentences do different work. A footer can satisfy one and leave the others open.
Article 50(4) is the deployer's duty. The second subparagraph is the text duty. The first covers a deep fake. Article 3(60) limits that term to AI-generated or manipulated image, audio or video that resembles existing persons, objects, places, entities or events and would falsely appear authentic or truthful. [5] Text sits outside the definition. A picture or a voice on the same page can meet it. Disclosure narrows when the deep fake is part of an evidently artistic, creative, satirical, fictional or analogous work, and a law-enforcement exception is stated. [1]
Article 50(1) sits on the provider. The system must tell a person, directly, that the interaction is with an AI system. Obviousness is an exit, judged for a reasonably well-informed, observant and circumspect person in the circumstances. So is the law-enforcement exception, which leaves out systems the public uses to report a crime. [1] Point 30's four elements are an AI system, an intention to interact, a direct interaction, and a natural person. The FAQ reads obviousness restrictively. It also attributes this design duty to Article 50(2). Article 50(1) and point 30 place it in Article 50(1), and that is the number used here. [2] [4]
Article 50(2) is the provider's mark. Synthetic audio, image, video or text is marked in a machine-readable format and made detectable, as far as technically feasible. Assistive standard editing, an alteration that leaves the input data and its semantics without a substantial change, and the criminal-offence authorisation sit outside the duty to the extent the statute states. [1] Point 71 limits the mark to a form software can identify, recognise and extract without human intervention. A perceptible label may sit beside it; that label complements the mark and leaves the marking duty in place. [2]
Article 50(5) requires the information in paragraphs 1 to 4 in a clear and distinguishable manner at the latest at first interaction or exposure, and in line with applicable accessibility rules. [1] Point 142 treats a manual, a buried menu, or unread terms of use as easy to miss. Point 144 says Article 50 adds no separate accessibility code. Where Directive 2016/2102 or Directive 2019/882 already applies, the Article 50 information meets that existing duty. [2]
Article 111(4) names providers, systems placed on the market before 2 August 2026, and Article 50(2). [3] Point 153 adds the words "or put into service." The statute used here says "placed on the market." [2] [3] The FAQ calls a limited grace period "envisaged," and only for that placement and that paragraph, with compliance from 2 December 2026. [4] Article 111(4) uses "shall." For a system that both interacts and generates, point 153 keeps the interaction disclosure on 2 August 2026 and parks the transition on the mark. [2] On 28 September 2026 the text duty and the interaction duty are already inside their window. The provider step still runs to 2 December 2026 for the systems Article 111(4) covers.
Article 1(40) of Regulation (EU) 2026/1744 moves Chapter III high-risk dates to 2 December 2027 for Annex III and 2 August 2028 for Annex I. Article 50 stays on 2 August 2026. [3] The high-risk timetable is a different chapter.
Guidelines point 152 sets the Article 50 maximum at EUR 15 000 000 or, for an undertaking, 3% of total worldwide annual turnover for the preceding financial year, whichever is higher. For SMEs, including start-ups, each fine is the percentage or the amount, whichever is lower. [2] The consolidated text of 27 July 2026 places the duty at Article 99(4)(g). Article 99(6) is that lower-of rule. Article 99(6a) extends it to small mid-cap enterprises for the fines in paragraphs 4 and 5. The EUR 35 000 000 or 7 percent band is Article 99(3), for Article 5 prohibitions. [3] [6] Point 152 also has the authority weigh the individual case. [2] The figure is a statutory maximum. This research retrieved no Article 50 decision imposing a fine. Enforcement remains unmeasured here.
A visible Article 50(4) label leaves the Article 50(2) mark unperformed. [1] [2]
The review record that lets a disease page skip the label
Both conditions have to be met. [1] Point 133 says so again. [2] A disease topic receives no class exemption. The diet article stays on the list of text that must be labelled unless the process is met.
Point 134 defines the first condition. Human review is a deliberate examination of the substance by one or more natural persons with relevant knowledge and professional judgement. The point's examples are academic peer review and professional validation chains. Fact-checking accuracy is a minimum that should be part of the review. Editorial control is control exercised in practice by a responsible editorial entity, the example being an editor-in-chief, with authority to approve, alter or reject on substantive grounds, including fact-checking and source trustworthiness. Existing media-service review procedures are left undisturbed. [2]
Point 135 excludes superficial checks. Spell-checking and grammatical correction are the examples, together with a paper policy, automated review, and cursory approval without substantive engagement. [2] The FAQ repeats the spell-checking line. [4] Point 136 covers the moment after sign-off. A substantive AI modification, supplement or reformulation after that sign-off voids the exception, and the resulting content is treated as AI-generated or manipulated. [2] The void attaches to the pass the later edit overtook. A fresh examination of the rewritten text is point 134 applied again. The guidelines describe no separate restart procedure.
Point 138 defines the second condition. A natural person, a legal person, an editorial board or the publishing company must hold ultimate legal responsibility for the publication, including the review. Identity and contact details should be public in an easily findable place where they are still unpublished. Online, terms and conditions or other user-facing legal information will do. Offline, a colophon or an edition notice will do. [2] That contact is accountability for the exemption. When a label is still required, point 142 still requires the artificial-origin disclosure to be perceivable at first exposure. A name on a legal page can satisfy the search for a responsible person while a buried label stays unseen. [2]
The guidelines' passing examples are processes, not topics: a newspaper article under an editor-in-chief, a peer-reviewed academic blog, an official's public-safety warning, a sustainability report reviewed by relevant staff, and a reviewed translation. [2] The failing examples are an unreviewed Union-policy site, a second model plus a human grammar check, and an unreviewed self-published book. [2] A health page inherits none of that as a class.
One disease-education page, as an illustration
Picture a public page, drafted by a generative system and opened on or after 2 August 2026 by people outside any closed list. It explains one disease and the difference between two care pathways. On the diet pattern it carries a text label unless the record below exists.
The guidelines describe the examination, the authority, the void and the contact. They leave the file format unstated. These rows are an illustration built from points 134, 135, 136 and 138, not a Commission template. The clinician, the rejected sentence and the legal-page contact are composite. No company is named. Nothing here was observed on a live site.
| What the record shows | Illustration that would meet the point | What fails the same point |
|---|---|---|
| Named reviewer, point 134 | A natural person is named, with relevant knowledge stated. Here, a clinician whose practice covers the disease, or a medical reviewer in a validation chain, examined the pathway comparison | A committee name or a job title, and nobody examined this page |
| Fact-check, point 134 | The record lists the checked statements, here the disease course and the pathway difference, and names the sources | A spell-check or a grammar pass, point 135's formal check |
| Authority to reject, point 134 | The person, or an editor-in-chief equivalent, can approve, alter or reject on the substance. Here the reviewer rejects one pathway sentence because the cited source fails to support it, and the rejection stays in the file | A policy that says clinicians review content, plus a cursory "approved" |
| After sign-off, point 136 | A sign-off time is recorded. After it, no generative system modifies, supplements or reformulates the substance | A model rewrites a clinical paragraph after the signature. The rewritten text needs its own examination |
| Editorial responsibility, point 138 | A person, board or publishing company holds ultimate legal responsibility, including for the review. Identity and contact are public. Here they sit on the legal page linked from the footer | A signature, and no public contact for the responsible person or function |
The committee's name is not the test. A medical, legal or regulatory review meets Article 50(4) when it is the point 134 examination and someone holds editorial responsibility under point 138. [2]
The Code of Practice draws a narrower paperwork line, and only for signatories. Media service providers who already work under editorial standards may rely on those procedures. Other signatories commit to a policy that names the responsible person, role and contact, and that overviews the measures and people used for the review. Commitment 4 releases those signatories from documenting each text. Unpublished contacts are to be published. [7] That is a voluntary policy. It leaves the requirement of an actual examination in place. Section 1 says adherence "does not constitute conclusive evidence of compliance" with Article 50(2) and (5). Section 2 uses the same clause for Article 50(4) and (5). [7] The code stops short of requiring this per-page file.
Five outputs, read against the quoted examples
A disease-education page
A public explainer follows the diet article. A one-line navigation label carries none of the facts. The long page is the one the record above is written for. [2] Point 154 leaves text published before 2 August 2026 free of a retroactive label, and it requires a label for earlier text published on or after that date when the text is in scope and the exemption is unmet. [2] The FAQ omits that publication-date split. [4] An image is separate. A synthetic patient who would appear real meets Article 3(60). A diagram that resembles no existing person or event sits outside it. [5]
A product page
Catalogue copy with no health, consumer-safety or sustainability claim is the out-of-scope example. A patient-effect, safety or environmental sentence is the parenthetical, so the three elements apply and public interest stays open. [2] Addleshaw Goddard says content relating to those claims may be public-interest information and may still need disclosure if it has not had substantive human review or editorial control. PETERKA PARTNERS says a different outcome is possible. [8] [9] "May" and "possible" are their words.
Point 116 treats colour correction, background replacement, arrangement and rescaling as likely minor for perceived authenticity. An AI product image that misleads on appearance, characteristics or use is on the deep-fake list. A real product on an AI background is on the other list when the advertisement is unlikely to mislead. [2] Points 117 and 132, below, are why the provider's mark leaves that disclosure unperformed. Annotations, boxes, contours, circles and heatmaps inside a medical device are on the Article 50(2) marking-exception list. [2] That line leaves the disease article, and a photorealistic render, alone.
A press release
The format goes unnamed. A wire release of investor information is a step past the corporate-report example, which is a report on the company site. A note to a named analyst, or an internal draft, follows the unpublished pattern. [2] A pure product description tracks the out-of-scope example. A disease explainer tracks the diet article. Point 12 asks for proportionate measures, contract terms or interface settings, so a partner's copy still shows the label at first exposure. [2] No wire service is named, and no contract term is invented. A synthetic executive or a cloned voice is a separate Article 3(60) question. The guidelines list a realistic chief-executive avatar as a deep fake. [2]
A site chatbot
A session-only answer follows the unpublished chatbot example, so no text duty arises. [2] Article 50(1) can still apply, and this piece treats no chatbot interface as obvious. [1] [2] Point 153 keeps that notice on 2 August 2026 even when the mark has the later date. [2] Paste the answer onto a public page and the session example stops.
A LinkedIn post or a newsletter
Subscriptions can still be published. An internal network stays unpublished. The guidelines print "indeterminate, fairly large" against "specific individuals" in a group "too small or insignificant," and they print no headcount. [2] The same disease explanation tracks the diet article on the public site and in a newsletter to that audience. On an internal network, and in a chatbot reply only the prompting person can see, publication stays unmet. A product description with no health, consumer-safety or sustainability claim stays the out-of-scope example on the page, in the newsletter and in the release. Adding the claim takes all three out together.
Forstemann's ordinary-marketing paragraph groups a newsletter with a product description and an ad, treats that group as outside public-interest text, and offers reading the piece and standing behind it as the review. The next paragraph separates an editorial piece on a matter of public interest. [10] The note sets no rule that every newsletter sits outside the duty. A disease newsletter is the editorial pattern, and "read and stand behind" falls short of point 134. Point 131 still tests the words that were sent.
Point 13 places a third-country deployer inside the transparency duties where Union dissemination is foreseen. Posting on the globally accessible internet is the illustration. Unforeseeable onward sharing, outside the deployer's control, is set aside. [2] Point 115, written for deep fakes, says a deployer need not assume broad public accessibility by default where the deep fake appears only on a subscriber page or in a corporate newsletter. That limit belongs to Article 3(60), and it leaves the text duty where point 131 put it. [2]
Who owes the machine-readable mark
Article 3(3) defines the provider as the person who develops an AI system, or has it developed, and places it on the market or puts it into service under that person's own name or trademark. Article 3(4) defines the deployer as the user under that person's authority, personal non-professional use aside. [5] Point 12 reads authority as responsibility for the decision to deploy and for how the system is used, outputs included. Technical control of the model is unnecessary. [2]
Point 11: a company that builds a chatbot in-house and puts it into service in the Union under its own name is the provider. It owes the Article 50(1) design duty, the Article 50(2) mark, and any Article 50(4) duty for what it publishes. [2] Point 15 allows both roles at once. An in-house system used for deep fakes owes the machine-readable mark and the perceivable label. [2] Buy a system another provider placed on the market under that provider's name, and the buyer is the deployer of what it publishes. The Article 50(1) design stays with the placing provider. Modify the system, including with new training data, and put the result into service under the buyer's name, and the buyer becomes provider of the new system while the original provider remains responsible for the initial one. [2] The contract, not the guidelines, sorts which of those facts is true.
Employees, and contractors under the same control, stay inside the legal person's deployer role. Point 14 names animators, web designers, content creators and journalists. A company that merely commissions an agency, with no decision and no control over the agency's use of AI, falls outside. Commissioning a page and choosing the system are different facts. [2] Hosts, platforms and broadcasters that only pass on third-party content fall outside too. Preserving marks is encouraged. Labelling alone does not make the labeller a deployer. [2]
The mark and the label fail in different directions. Point 71 defines the mark as something software extracts without a person. A readable footer can perform the deployer disclosure and still leave the mark unperformed.
Point 117, written for deep fakes, bars a deployer from treating the provider's embedded mark as the disclosure. Those marks are built for software. They leave the viewer short of a disclosure that is immediately clear and distinguishable, and they call for a technical tool. [2] Point 132 carries the same standard to Article 50(4) text: visible or audible measures, perceivable by a person, with no need for a technical tool and no need for a dedicated action. [2] A watermark software can extract can perform the mark and still leave the person uninformed. That is the point 117 result for a deep fake and the point 132 result for text.
Spell-checking shows why the paragraphs stay apart. Point 90 treats standard editing as small changes for readability, grammar, quality and format that generate no new content. Material changes to meaning, style or intent fall outside it, and so do summaries and paraphrases that change style, structure and meaning. [2] The same spell-check can sit inside the Article 50(2) exception and still fail point 135. A published disease-page summary is also the diet pattern.
The code is voluntary. The Commission policy page says the Commission and the AI Board have confirmed it as an adequate voluntary tool to demonstrate compliance. [11] The conclusive-evidence sentence above limits what that confirmation does. It supplies no share of life-sciences companies and proves nothing about a given page. For text the code still treats as published, a signatory uses an icon or equivalent whose main visual element is the capitalised acronym "AI". Sub-measure 1.2.2(f) places it, for example, above or at the top, near the headline, or in the colophon at the beginning, and allows a label on only the AI-generated or manipulated part. [7] The icon performs a disclosure. It leaves the machine-readable mark untouched.
The operative Article 50(7), inserted by Article 1(20) of Regulation (EU) 2026/1744, has the Commission encourage codes, assess adequacy while taking utmost account of the Board's opinion, and allows an implementing act if the Commission deems a code inadequate. [3] The AI Act Service Desk still shows a pre-amendment paragraph 7. Paragraphs 1 to 6 on that page are the text used here. Paragraph 7 on that page is set aside. [1] [3]
The five-output table
A cell that repeats a Commission sentence is a quotation. Any other cell applies point 131, Article 50(1), Article 50(2) or Article 3(60) to an output the guidelines leave unnamed. The table sorts this desk's five outputs by applying the quoted tests; the Commission has published no classification of life-sciences page types.
| Output | Text label (Art. 50(4)) | What lets you skip it | Other Article 50 duties (provider notice, provider mark, deep fake) | Basis |
|---|---|---|---|---|
| Disease-education page | In scope on the diet example when the page informs an indeterminate public about a disease. A one-line label that communicates no such facts falls outside point 131 | Both conditions: a named person examines and fact-checks the substance and can reject it, and a public contact holds editorial responsibility. A substantive AI edit after sign-off voids that pass | A static page needs no interaction notice. The provider still owes the mark on generated text. Image, audio or video is a separate Article 3(60) question | Diet article: quotation. This page: application of point 131 |
| Product page | A description or advertisement with no health, consumer-safety or sustainability claim is the out-of-scope example. A page that includes such a claim leaves that example and is tested on the three elements | The exemption matters once the text is inside the duty. Copy that stays inside the out-of-scope example has no Article 50(4) text label to skip | A static page needs no interaction notice. An on-page assistant is the chatbot row. The provider owes the mark unless standard editing applies. A misleading product image can be a deep fake | Product-description sentence: quotation. Health-claim page: application |
| Press release | Unnamed. Investor information for an indeterminate public tracks the corporate-report example. A disease explainer tracks the diet example. A pure product description without those claims tracks the out-of-scope example | The same two conditions, once the text is in scope. Point 12: proportionate measures so the disclosure stays perceivable at first exposure after partners distribute it | A release needs no interaction notice. The provider owes the mark on generated or substantially altered text. A synthetic executive image, voice or video is a separate Article 3(60) question | Application, tied to the corporate-report, diet or product example |
| Site chatbot | An answer shown only to the person who prompted it is the unpublished chatbot example, so no text duty arises. A posted transcript is tested on topic | A session-only answer needs no text-duty exemption. A posted answer that is in scope needs both conditions in points 134 and 138 | Provider interaction notice from 2 August 2026. Article 111(4) moves only the mark, to 2 December 2026, for pre-August systems. A text reply falls outside Article 3(60) | Session-only access: quotation. A public transcript is an application |
| LinkedIn post or newsletter | Subscriptions can still be published. An internal network falls outside publication. A disease newsletter to an indeterminate, fairly large audience tracks the diet example. A product post without those claims tracks the product example | Point 134 is the test. Reading the piece and standing behind it falls short of that examination. The channel leaves the test unchanged | A post needs an interaction notice only when it is itself a direct AI interaction. Point 13 reaches a third-country deployer who foresees Union use. Point 115 limits the deep-fake audience alone | Publication and subscription sentences: quotations. Disease newsletter and product post: applications |
Open one disease URL, one product URL and one chatbot transcript. Exempt the disease URL only with the worked record, including a named person who can reject the substance and no substantive AI edit after sign-off. Keep the product URL outside the text duty only while the copy stays inside the quoted description. Keep the transcript outside that duty while only the prompting person can see it. The interaction notice and the mark stay separate.
What these texts leave to other rules
Article 50(6) says paragraphs 1 to 4 shall not affect Chapter III, and shall be without prejudice to other Union or national transparency obligations for deployers. [1] The code lists the same parallel operation: data protection, consumer protection, digital services, intellectual property, media law, political advertising, criminal law and other frameworks. [7] Directive 2001/83/EC, Regulation (EU) 2017/745 and Regulation (EU) 2017/746 go unnamed in these passages as the way to decide an Article 50(4) label. Overlap with medicinal-product advertising rules, or with medical-device and in-vitro-diagnostic claim rules, stays open. No national life-sciences guidance applying Article 50(4) to these pages was retrieved. That gap is a limit of this search rather than a verified absence.
Point 5 says the guidelines are non-binding, and that any authoritative interpretation of the AI Act may ultimately only be given by the Court of Justice of the European Union. [2] Communication C(2026) 5054 of 20 July 2026 says the guidelines will be formally adopted when all language versions are available, and that they will be applicable only from that moment. [12] The Commission library page, publication date 20 July 2026 and last update 31 July 2026, says the Commission adopted these guidelines. [13] No later adopting act was retrieved. Article 50's application on 2 August 2026 stands in the regulation, independent of that later adoption. The duties in the table are the duties in the regulation. The examples are the guidelines' examples.
Member State market-surveillance authorities supervise Article 50. The AI Office has the limited role point 151 describes for systems built on general-purpose models in the cases that point lists. The European Data Protection Supervisor supervises Union institutions, bodies and agencies. [2] A commentary trail is not an official decision. No Article 50 measure imposed by the Commission or by a national market-surveillance authority was retrieved. That result is a gap in this pass. It measures neither enforcement nor whether any authority has acted.
An Article 50(4) disclosure tells the reader the text was artificially generated or manipulated. It does not clear a promotional claim, a device claim or a medicine claim.
VayoMed publishes through the client's review path. That path is not, by itself, the Article 50(4) exemption, and an Article 50 label leaves the client's medical, legal or regulatory review in place. In that workflow the operating practice is a named reviewer with authority to reject the substance, and a public contact for the person who holds editorial responsibility. That practice states no compliance result for any page.
Sources
1. European Commission, AI Act Service Desk, "Article 50: Transparency obligations for providers and deployers of certain AI systems," Regulation (EU) 2024/1689, paragraphs 1 to 6. Paragraph 7 on this page is the pre-amendment text and is not cited as current law. Accessed 28 September 2026.
2. European Commission, guidelines on the implementation of the transparency obligations for certain AI systems under Article 50 of Regulation (EU) 2024/1689, annex to Communication C(2026) 5054, 20 July 2026, points 5, 11 to 16, 30, 71, 90, 114 to 117, 131 to 138, 142, 144, 151 to 154. Accessed 28 September 2026.
3. Regulation (EU) 2026/1744 of the European Parliament and of the Council of 8 July 2026 amending Regulation (EU) 2024/1689 (Digital Omnibus on AI), OJ L, 24 July 2026, Article 1(20), Article 1(38), Article 1(39)(b), Article 1(40) and Article 4. Accessed 28 September 2026.
4. European Commission, "Transparency obligations under Article 50 of the AI Act," frequently asked questions, last update 24 July 2026. Accessed 28 September 2026.
5. European Commission, AI Act Service Desk, "Article 3: Definitions," Regulation (EU) 2024/1689, points (1), (3), (4) and (60). Accessed 28 September 2026.
6. Consolidated text of Regulation (EU) 2024/1689 of 27 July 2026, Article 99(3), (4)(g), (6) and (6a). Accessed 28 September 2026.
7. European Commission, Code of Practice on Transparency of AI-generated Content, 10 June 2026. Section 1, objective (a), on adherence and Article 50(2) and (5). Section 2, objective (a), on adherence and Article 50(4) and (5); the definition of published text; paragraph (f) on other Union and national law; Measure 1.1; Sub-measure 1.2.2(f); Commitment 4. Accessed 28 September 2026.
8. Addleshaw Goddard, "AI transparency under the AI Act: what businesses need to know before 2 August 2026," section "AI-generated text." Accessed 28 September 2026.
9. PETERKA PARTNERS, "Article 50 of the AI Act in practice: transparency of AI systems, content labeling and deepfakes," sections on what "published" means and on professional and corporate communications. Accessed 28 September 2026.
10. Forstemann, "Labelling AI content: what Article 50 requires and what it does not," section "What does this mean for ordinary marketing?" Accessed 28 September 2026.
11. European Commission, "Code of Practice on Transparency of AI-generated Content," policy page, statement that the Commission and the AI Board have confirmed the code as an adequate voluntary tool. Accessed 28 September 2026.
12. European Commission, Communication to the Commission C(2026) 5054, 20 July 2026, approval of the content of the draft Article 50 guidelines, paragraph beginning "Such clarity is essential." Accessed 28 September 2026.
13. European Commission, "Guidelines on transparency obligations for providers and deployers of AI systems," library page, publication 20 July 2026, last update 31 July 2026. Accessed 28 September 2026.
Founder @ VayoMed, RAC
DJ is a Regulatory Affairs Certified (RAC) professional with deep expertise in life sciences go-to-market strategy. He helps medical device and healthcare companies navigate the intersection of regulatory compliance and digital visibility, ensuring brands are positioned for success in both traditional and AI-powered search environments.
Need the engine that comes after clearance?
VayoMed builds and runs your website, content, LinkedIn, domains, PR, and AI visibility as one fully managed $24K/year annual subscription.