A US pharmaceutical or device company that puts a chatbot on a patient-support site, lets a generative tool draft disease-awareness copy or builds a campaign around synthetic images may approach AI communications through familiar questions about truthfulness, balance and consistency with approved product labeling. The FDA's prescription-drug advertising rules address content and presentation, including fair balance and misleading claims. California's bot statute reaches only a bot used online with intent to mislead a person about its artificial identity and knowingly deceive that person about the content of the communication, in order to incentivize a sale or influence a vote, and a clear, conspicuous disclosure can exclude liability under that section, while the state's health-care disclosure rule binds health facilities, clinics and physician practices rather than manufacturers and does not apply to a message read and reviewed by a human licensed or certified health care provider.121 C.F.R. § 202.1; Cal. Bus. & Prof. Code § 17941; Cal. Health & Safety Code § 1339.75. Since 2 August 2026 the European Union asks a different question. Art. 50 AI Act sets distinct duties for direct-interaction notices, machine-readable output marks and perceptible disclosure of deep fakes and certain published text, each subject to its own scope and exceptions. Breaches engage administrative fine ceilings that depend in part on whether the offender is a smaller business.2Regulation (EU) 2024/1689 (Artificial Intelligence Act) [2024] OJ L 2024/1689, Art. 3(3) and (4), Art. 50 and Art. 99(4)(g), (6), (6a) and (7). For a life-sciences company the new layer does not replace the rules that already govern what may be said about a medicine or a device. It sits on top of them, and it allocates its duties by a test that the company's vendor contracts were not written against.
1. Who Owes the Chatbot Disclosure When a Pharma Brand Runs a Vendor's AI?
Art. 50(1) AI Act is addressed to the provider, not to whoever runs the conversation. The provider must ensure that a system intended to interact directly with natural persons is designed and developed so that the persons concerned are informed that they are interacting with an AI system, unless this is obvious to a person who is reasonably well-informed, observant and circumspect, taking into account the circumstances and the context of use. Art. 3(3) AI Act makes the provider whoever develops an AI system, or has one developed, and places it on the market or puts it into service under its own name or trademark, and Art. 3(4) AI Act makes the deployer whoever uses a system under its authority.2 The Commission's non-binding guidelines on Art. 50 AI Act, whose content the Commission approved on 20 July 2026 while deferring their formal adoption until every language version was available, treat a company that builds a chatbot in-house and puts it into service under its own name as a provider. They treat a company that takes an existing generative system, modifies it, for instance with new training data, and puts it into service under its own name as the provider of a new system. Authority, on the deployer's side, means responsibility for the decision to deploy and for the manner of use, and does not necessarily require technical control.3Commission Guidelines on the transparency obligations under Art. 50 AI Act, C(2026) 5054 final (20 July 2026), content approved; Commission Q&A (updated 24 July 2026).
That test sorts the typical pharma deployment poorly. A patient-support assistant carrying the product's name, built on a vendor's conversational platform, grounded in the summary of product characteristics through retrieval and shaped by the brand team's instructions, can be described as the vendor's system in the pharma company's use or as the pharma company's system developed by a vendor. The two descriptions send the design duty of Art. 50(1) AI Act, and the marking duty of Art. 50(2) AI Act examined below, to different companies. The services agreement will call the pharma company a customer or a licensee. An authority will ask the questions Art. 3(3) AI Act asks, under whose name the system was put into service and who had it developed, and the answers are facts about branding and about the build that no allocation clause changes. A vendor that has contracted to handle AI Act compliance may not be the provider. A brand team that believed it was buying a service may be.
The vendor contract decides which company does the work of Art. 50 AI Act. Art. 3(3) AI Act decides which company owes the duty, and in a branded pharma chatbot the two need not be the same company.
The exception for obvious interaction does less work in health care than in most sectors. The guidelines interpret it restrictively and confine it to cases in which almost no doubt about the nature of the interaction is left for an average member of the targeted and reasonably foreseeable audience. They list health advice among the contexts in which a single notice may not suffice and periodic reminders are likely to be necessary, and they treat a disclosure found only in terms and conditions, a generic reference to an assistant, or a human-like persona as inadequate on its own. Among their examples, an interactive system intended only for properly trained health professionals to support diagnosis and suggest treatment is listed as an obvious interaction, while a helpdesk chatbot whose replies users may take for a human's is not.3 Art. 50(5) AI Act requires the information under its first four paragraphs to be clear and distinguishable, provided by the first interaction or exposure, and compliant with applicable accessibility requirements. A voice interaction therefore needs perceptible disclosure, not merely an unseen screen label. A medical-information assistant for prescribers sits between the two. Its intended audience is professional, but the guidelines measure obviousness against the reasonably foreseeable audience as well as the intended one, and a public product website rarely admits prescribers alone. AI agents raise the requirement again. The guidelines expect an agent that writes to a person to disclose both its artificial nature and the party on whose behalf it acts, and an agent that drafts and sends emails to physicians for a field team falls within that description.
Compared with California's bot rule in section 17941 of the Business and Professions Code, Art. 50(1) AI Act does not depend on an intent to mislead. The California statute makes unlawful the use of a bot with intent to mislead in a commercial transaction or an election, and a disclosure takes the operator outside it. Art. 50(1) AI Act is a design obligation, breached where a required notice is missing, irrespective of that intent. This contrast does not describe every US disclosure law: Utah's generative-AI provisions, for example, require disclosure in specified consumer and regulated-service interactions without that intent test.1 The EU rule also reaches systems operated from abroad. Under Art. 2(1)(a) and (c) AI Act the Regulation reaches providers placing systems on the Union market or putting them into service there, wherever they are established, and providers and deployers in third countries where the output of the system is used in the Union. The duty to appoint an authorized representative established in the Union, in Art. 22 AI Act, attaches, among AI systems, only to high-risk ones.2 A US parent that operates the conversational layer of a global patient-support program as its provider therefore does not answer for Art. 50 AI Act through a representative in the Union. It owes the obligation itself.
One consequence of the allocation is easy to miss. The answer a chatbot gives one patient is not published text under the second subparagraph of Art. 50(4) AI Act, because the guidelines place text available only to the user who prompted it outside that provision. The conversation is therefore governed by the provider's duties to disclose and to mark, not by the deployer's duty to label. If the pharma company is only a deployer, the AI Act places the design of the disclosure and the marking of outputs within Art. 50(2) AI Act with the provider, and the pharma company's influence over either is whatever its contract secures.
2. Which AI Outputs Need Marks, and Who Has Until 2 December 2026?
Art. 50(2) AI Act requires providers of AI systems, including general-purpose AI systems, that generate synthetic audio, image, video or text content to ensure that the outputs are marked in a machine-readable format and detectable as artificially generated or manipulated, through technical solutions that are effective, interoperable, robust and reliable as far as this is technically feasible. The duty does not apply to the extent that a system performs an assistive function for standard editing or does not substantially alter the input data or its semantics.2 The guidelines draw the lines where regulated content lives. Grammar correction and AI-generated translation count as standard editing, while summaries and rewriting that change style, structure or meaning do not. Technical processing and reconstruction of medical images within devices is listed as standard editing. Output that merely reproduces existing content, or presents and arranges it, falls outside the provision altogether.3 A retrieval system that returns a passage of the approved package leaflet word for word and a generative system that explains the same passage in its own words therefore sit on opposite sides of Art. 50(2) AI Act, although to the patient reading the screen they may look the same.
The Digital Omnibus on AI added transitional relief for systems placed on the market before 2 August 2026 and rewrote the paragraph on codes of practice, but left the duties in Art. 50(1) to (6) AI Act as they were. Regulation (EU) 2026/1744 was adopted when the Council approved the European Parliament's position on 29 June 2026, was signed on 8 July 2026, appeared in the Official Journal on 24 July 2026 and entered into force three days later, on 27 July 2026, six days before Art. 50 AI Act began to apply.4Regulation (EU) 2026/1744 (Digital Omnibus on AI) [2026] OJ L 2026/1744, Art. 1, point (39)(b), inserting Art. 111(4) AI Act; Recital 38; in force 27 July 2026. It inserted Art. 111(4) AI Act, under which providers of generative systems placed on the market before 2 August 2026 must bring them into line with Art. 50(2) AI Act by 2 December 2026, and Recital 38 of Regulation (EU) 2026/1744 explains the four months as time to adapt practices without disrupting the market. The Commission's answers to frequently asked questions and the guidelines both confine the transition to marking and detection. Only a generative system placed on the market before 2 August 2026 benefits from the Art. 111(4) AI Act marking transition to 2 December 2026; its applicable chatbot disclosure and deployer-label duties are not deferred. For other systems, the marking duty applies from 2 August 2026. Outputs and deep fakes generated before 2 August 2026 need not be marked or labeled after the fact, but text generated before that date and published after it must carry the label unless an exception applies.3
How the marks are made is left to the state of the art and, in practice, to a voluntary code. The Code of Practice on Transparency of AI-generated Content was published on 10 June 2026, and on 8 July 2026 the Commission found that it adequately covers the obligations in Art. 50(2), (4) and (5) AI Act, with the AI Board adopting its own adequacy assessment the following day.5Code of Practice on Transparency of AI-generated Content (10 June 2026); Commission Opinion C(2026) 4839 final (8 July 2026); AI Board adequacy assessment (9 July 2026). The guidelines say that authorities will focus their supervision of signatories on adherence to the code, and that providers and deployers who do not sign should expect more requests for information and are expected to show, for instance through a gap analysis against the code, how they comply. The Commission's opinion adds the qualification a regulated company will notice first: adherence is not conclusive evidence of compliance. For a pharma company that has concluded it is the provider of its branded assistant, the decision whether to sign, and the question whether its vendor's watermarking at the level of the model reaches an obligation the Regulation imposes at the level of the system, remain separate compliance questions. The December transition is relevant only if the system was placed on the market before 2 August 2026.
For a device manufacturer the calendar holds an inversion that is easy to read the wrong way. Software that is an AI system and a device requiring third-party conformity assessment is high-risk through Art. 6(1), Annex I AI Act, and after the Digital Omnibus the high-risk obligations reach it only from 2 August 2028 under Art. 113 AI Act. The transparency duties do not wait for them. Art. 50(6) AI Act provides that its first four paragraphs do not affect the high-risk requirements, the guidelines treat the two sets of obligations as cumulative, and the grandfathering rule of Art. 111(2) AI Act for high-risk systems already on the market does not extend to Art. 50 AI Act.3 A conversational coaching function in a connected device therefore carries the disclosure duty from 2 August 2026, and a generative feature that writes a patient-readable summary of the device's readings carries the marking duty from that date or, if the system was placed on the market before 2 August 2026, from 2 December 2026, up to two years before the high-risk rules reach either. The interaction between the AI Act and the device conformity assessment is examined in the analysis of medical AI under the AI Act and MDR. These transparency duties precede the deferred high-risk requirements; other AI Act provisions, including AI literacy and prohibited practices, already apply.
Scroll the diagram horizontally to read all three panels.
3. Does MLR Review Count as Editorial Control Over AI-Written Health Text?
The deployer's labeling duty for text is the one closest to the daily work of a medical affairs or communications team. Under the second subparagraph of Art. 50(4) AI Act, a deployer of an AI system that generates or manipulates text published with the purpose of informing the public on matters of public interest must disclose that the text has been artificially generated or manipulated, unless the AI-generated content has undergone a process of human review or editorial control and a natural or legal person holds editorial responsibility for its publication.2 The guidelines, and the Commission's answers to frequently asked questions, list public health among the matters of public interest, beside consumer safety and scientific developments that may be a relevant subject of public debate. Their examples bring health content in directly. AI-manipulated parts of a lifestyle article comparing the effects of diets on a disease are within the provision, while their example of AI-manipulated advertising or product-description text that falls outside it expressly excludes claims related to health, consumer safety or sustainability.3 A disease-awareness page, a patient brochure posted online, a press release on trial results or a medical-information article on a public site written with a generative tool is, on that reading, a candidate for the label. The carve-out for advertising copy is drawn so as not to cover copy that makes health claims, which describes most of what a pharmaceutical or device company says about its products.
Two threshold questions decide whether the provision reaches a given piece at all. The first is publication. The guidelines treat text as published when it is accessible to an indeterminate, fairly large number of unrelated readers, and as unpublished when access is restricted to a closed private group or the text is internal to an organization. Prescription-medicine promotion in the Union is confined to persons qualified to prescribe or supply, and online professional material is commonly placed behind registration. Whether a password-protected portal open to every registered physician in a country is a closed group or an indeterminate public is not a question the guidelines answer. The second is purpose. The text must aim to communicate knowledge, opinions or facts to the public, and a company's own characterization of a piece as promotional rather than informative is not obviously the test an authority will apply.
The exception is where medical, legal and regulatory review appears to fit, and where the fit is less exact than a promotional-review team may assume. Human review and editorial control are alternative routes, each requiring editorial responsibility for publication. The guidelines define human review as the deliberate examination of the substance of the content by persons possessing relevant knowledge and professional judgment, name professional validation chains as an example, and make fact-checking of accuracy a minimum. Editorial control instead concerns an editorial entity's practical authority to approve, alter or reject the substance of the text, including fact-checking and source assessment. The guidelines exclude superficial, formal or procedural checks, the mere existence of an editorial policy, automated review and cursory approval without substantive engagement. An MLR process that tests every claim against its reference, under a named approver, resembles a professional validation chain, and the guidelines' own example of a sustainability report reviewed by professionals in the compliance function points the same way. Three features of common practice sit less comfortably. The guidelines provide that any substantive AI intervention after sign-off voids the exception, so substantive personalization or rewriting cannot rely on approval of an earlier version. The resulting text may need a fresh qualifying review or editorial-control process; purely technical formatting or unchanged assembly is not automatically a substantive intervention. Their example of AI-supported translation of a human-written article qualifies only because the translation was reviewed, which sits oddly with their treatment of AI translation as standard editing for the marking duty, and not every localization workflow reviews each language version. And editorial responsibility requires a natural or legal person holding the ultimate legal responsibility for publication, whose identity and contact details the guidelines say should be publicly available in an easily findable place, a disclosure that promotional-review workflows were not designed to make.3
The US analogue the reader will reach for is California's health-care rule, which exempts a covered communication read and reviewed by a human licensed or certified health care provider. The EU exception asks more, and asks it of a different population of text. Its human-review route turns on relevant knowledge and professional judgment rather than professional licensure, while its alternative editorial-control route concerns authority over the substance of the text. Both require editorial responsibility, which the guidelines expect to be publicly identifiable; California's quoted exception does not impose that condition.1
Synthetic images and video in campaigns fall under the first subparagraph of Art. 50(4) AI Act, and the definition in Art. 3(60) AI Act is broader than its name suggests. A deep fake is 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. The guidelines read existing as including subjects that can plausibly exist, count realistic AI-generated avatars and personas as persons and the depiction of professional services as events, and give as examples an AI-generated celebrity influencer in an advertisement and an AI-generated product image that misleads about the product; mice arguing about cheese in an advertisement are not a deep fake. A photorealistic patient describing relief from symptoms, or a synthetic physician explaining a mechanism of action, can on that reading be a deep fake if, in its context and for its foreseeable audience, it would falsely appear authentic or truthful. Photorealism alone does not decide that question. Where the definition is met, the deployer must disclose the artificial origin in a way a person can perceive without technical tools, because the machine-readable mark placed under Art. 50(2) AI Act does not discharge the label. The deployer is not necessarily the marketing authorization holder. The guidelines say that a company that merely commissions an advertising agency, without deciding or controlling whether and how the agency uses AI, is not a deployer.3 A holder that keeps its distance from the agency's tools may stay outside the AI Act duty and keep, under the medicines advertising rules examined next, its own duty to ensure that the campaign conforms.
4. The Label Does Not Cure the Content: Advertising Law and the Enforcement Map
Art. 50(6) AI Act states that its first four paragraphs are without prejudice to other transparency obligations that Union or national law lays down for deployers, and Recital 137 AI Act adds that compliance with the transparency obligations is not to be read as indicating that the use of the system or its output is lawful under the AI Act or under other Union or Member State law.2 The guidelines draw the practical consequence: a label does not permit the generation and dissemination of a deep fake that is unlawful under other law, with misleading advertising among their examples.3 For a medicines company the other law includes Titles VIII and VIIIa of Directive 2001/83/EC, examined in the analysis of Swiss and EU pharmaceutical advertising. Art. 86(1) of the Directive defines advertising broadly enough to include any form of door-to-door information, canvassing activity or inducement designed to promote the prescription, supply, sale or consumption of medicinal products. Art. 88(1)(a) of the Directive requires the Member States to prohibit advertising prescription-only medicines to the general public, and Art. 90 of the Directive lists what public advertising may not contain, among it material that gives the impression that a medical consultation is unnecessary, in particular by offering a diagnosis, material that refers to a recommendation by scientists, health professionals or persons whose celebrity could encourage consumption, and improper, alarming or misleading pictorial representations of the action of a medicine on the body.6Directive 2001/83/EC [2001] OJ L311/67, Titles VIII and VIIIa: Art. 86, Art. 88(1)(a), Art. 90, Art. 97 and Art. 98.
The question a generative assistant raises under those provisions is older than the AI Act and sharper because of it. Art. 86(2) of the Directive excludes from advertising the correspondence needed to answer a specific question about a particular medicinal product, and information relating to human health or diseases provided that there is no reference, even indirect, to medicinal products. In MSD Sharp & Dohme v Merckle the Court of Justice held that Art. 88(1)(a) of the Directive does not prohibit a pharmaceutical company from disseminating on its website, to someone who seeks it, the faithful reproduction of the packaging and the literal and complete reproduction of the approved package leaflet or summary of product characteristics, but that information which has been selected or rewritten by the manufacturer, and which can be explained only by an advertising purpose, is prohibited.7Case C-316/09 MSD Sharp & Dohme v Merckle ECLI:EU:C:2011:275, operative part. A generative system selects and rewrites by construction. Whether a patient-support assistant that paraphrases the leaflet in answer to a question is correspondence answering a specific question, a website making approved information available to someone who seeks it, or selected and rewritten information that the Court's formula treats as advertising, is a question a 2011 judgment could not have asked. For a prescription-only product the answer decides whether a product-specific assistant can face the public at all. The AI Act points in the same direction from the other side, since output that merely reproduces existing content falls outside the marking duty while a paraphrase falls within it.
The Directive also assumes a world of discrete advertisements. Art. 98 of the Directive requires the marketing authorization holder to establish a scientific service in charge of information about its products, to keep available a sample of all advertisements emanating from its undertaking with a statement of their addressees, method of dissemination and date of first dissemination, and to ensure that the advertising of its undertaking conforms to Title VIIIa.6 What counts as a sample of an advertisement when a system produces a different text for each conversation, and whether the holder's duty to ensure conformity reaches an agency's generative tool that, on the guidelines' reading, the holder does not deploy, are questions that national authorities may meet before any court does.
National law adds detail. In Germany, § 10 Abs. 1 HWG confines advertising for prescription-only medicines to physicians, dentists, veterinarians, pharmacists and persons lawfully trading in them, § 3 HWG prohibits misleading advertising, and § 11 Abs. 1 HWG forbids advertising outside the professional circles that refers to recommendations by scientists, health professionals or well-known persons, uses pictorial representations of the effect of a medicine on the body in an abusive, repugnant or misleading way, or uses third-party statements in such a way, with the prohibition on third-party statements and several others extended to medical devices.8Heilmittelwerbegesetz (HWG), § 3, § 10 Abs. 1 and § 11 Abs. 1. A synthetic physician recommending a medicine in a consumer campaign engages § 11 Abs. 1 Satz 1 Nr. 2 HWG whether or not it carries a label. The HWG is enforced in Germany not only by the authorities but by competitors and associations, through § 3a UWG, which treats as unfair the breach of a statutory provision that is also intended to regulate market conduct in the interest of market participants where the breach is liable to affect their interests appreciably, and through the standing that § 8 Abs. 3 UWG gives them.9Gesetz gegen den unlauteren Wettbewerb (UWG), § 3a and § 8 Abs. 3. Whether Art. 50 AI Act is such a provision, so that a missing chatbot notice or deep-fake label becomes a matter a competitor can pursue by injunction, is a question the Regulation does not answer. An affirmative answer would add a private enforcement channel to a duty the Regulation assigns to public authorities.
A device manufacturer carries a parallel constraint in Art. 7 MDR, which prohibits, in labeling, instructions for use, making available, putting into service and advertising, text or pictures that may mislead the user or the patient about the device's intended purpose, safety and performance, one of the four means it lists being the suggestion of uses other than those for which the conformity assessment was carried out.10Regulation (EU) 2017/745 (MDR) [2017] OJ L117/1, Art. 7. A generative assistant that recommends a use outside the conformity-assessed intended purpose can engage Art. 7(d) MDR where the statement may mislead the user or patient. Merely receiving or answering a question about such a use does not establish that condition: a refusal or neutral explanation need not suggest the use. An AI notice does not cure a misleading recommendation.
Switzerland has its own advertising rules and no equivalent general AI transparency regime. Switzerland is not an EU Member State and has no counterpart to Art. 50 AI Act. On 12 February 2025 the Federal Council chose to ratify the Council of Europe's AI convention and to make the legislative amendments this requires sector by sector where possible, confining cross-sector rules to core areas of fundamental rights such as data protection, with a consultation draft on transparency, data protection, non-discrimination and oversight due by the end of 2026, a path examined in the analysis of Swiss AI regulation compared with the EU AI Act.11Federal Council, media release of 12 February 2025; BAKOM, consultation draft due by the end of 2026. The Swiss advertising ordinance already reaches much of what a synthetic campaign might do. Art. 16(3) AWV requires public advertising to be recognizable as such, Art. 22(a) AWV bars advertising that makes a medical examination appear unnecessary, in particular by offering a diagnosis, and Art. 22(h) AWV bars public advertising that shows persons in the professional clothing of medical personnel, a prohibition that does not ask whether the physician on screen exists.12Arzneimittel-Werbeverordnung (AWV) (SR 812.212.5), Art. 16(3) and Art. 22(a) and (h). A Swiss company whose chatbot or campaign output is used in the Union is, for its part, within Art. 2(1)(c) AI Act.
Who Enforces the Label
National authorities are central to Art. 50 AI Act enforcement, subject to the AI Office allocation discussed below. Art. 99(4)(g) AI Act sets a ceiling of EUR 15 million or, for an undertaking, 3% of its total worldwide annual turnover for the preceding financial year, whichever is higher. Under Art. 99(6) and (6a) AI Act, the lower of those limits applies to small and medium-sized enterprises, including start-ups, and small mid-cap enterprises. The actual fine depends on the circumstances listed in Art. 99(7) AI Act, rather than on turnover alone. Art. 70 AI Act left the designation of market surveillance authorities to the Member States under a governance chapter that has applied since 2 August 2025. Any natural or legal person who considers the Regulation infringed may complain to the relevant authority under Art. 85 AI Act.2 Germany's implementing act was promulgated on 28 July 2026 and entered into force on 29 July 2026, four days before Art. 50 AI Act applied. Under § 2 Abs. 1 KI-MIG the Bundesnetzagentur is the market surveillance authority unless the Act provides otherwise, and under § 2 Abs. 2 KI-MIG the authorities designated to enforce the Union harmonization legislation in Section A of Annex I AI Act, which lists the MDR, also act as market surveillance authorities for AI systems connected with products to which that legislation applies.13KI-MIG (BGBl. 2026 I Nr. 223), § 2 Abs. 1 and 2; in force 29 July 2026. For medical devices those are, under § 85 Abs. 1 MPDG, the authorities of the Länder, which under § 77 Abs. 2 MPDG already verify compliance with the rules on healthcare advertising for devices.14Medizinprodukterecht-Durchführungsgesetz (MPDG), § 77 Abs. 2 and § 85 Abs. 1. A device company's chatbot and what it says may therefore meet one authority of a Land under two regimes, while a pharma company's patient-support assistant answers to the Bundesnetzagentur for its disclosure and to other bodies for its content.
Where a conversational system is based on a general-purpose AI model developed by the same provider, or by a provider in the same undertaking, Art. 75(1) AI Act, as amended by the Digital Omnibus, makes the AI Office exclusively competent for that provider's obligations, excluding AI systems related to products under Annex I AI Act and reaching a deployer only where it is also the provider or part of the same undertaking.2
5. What the Duties Mean for US Pharma and MedTech Companies
The statutory role and the contractual allocation can diverge in the same program. A pharma company may commission and brand the conversational system, leave technical operation to a vendor, and keep approval of its communications in-house. Art. 3(3) AI Act examines development and placement on the market or putting into service under the company's name; neither a vendor's compliance promise nor a marketing team's description of its role settles that test. The applicable marking date then depends on whether the system was placed on the market before 2 August 2026.
Review introduces a separate issue. An approved source document and the text ultimately published may differ after substantive personalization or rewriting, while a private answer and a publicly available health article engage different parts of Art. 50 AI Act. Human review or editorial control can support the text-label exception only with editorial responsibility for publication. Neither route removes medicines or device advertising restrictions. The marketing authorization holder may retain advertising obligations even where an agency, rather than the holder, is the AI Act deployer.
For a US-headquartered group, the financial and communications consequences extend beyond the local campaign. Art. 99 AI Act uses worldwide turnover in setting fine ceilings, subject to the lower-cap rules for SMEs and small mid-cap enterprises; it does not prescribe an automatic turnover-based penalty. A finding about undisclosed AI interactions or synthetic testimonials may also matter to risk disclosures and insurance coverage, depending on materiality and policy terms. California's specified rules and other US state laws use different triggers. A common global interface therefore does not establish that the legal conditions are common.