Does AI-written B2B content need a disclosure label in the EU?

Article 50 of Regulation (EU) 2024/1689 became applicable on 2 August 2026, and from that date a breach of it is fineable. The Digital Omnibus did not postpone it. That instrument moved the Annex III high-risk deadline to 2 December 2027 and touched Article 50 only narrowly: the Commission describes "a targeted grandfathering rule only with regard to the marking and detection obligations under Article 50(2)" for generative systems placed on the market before 2 August 2026, giving those providers until 2 December 2026. The duty to disclose an AI interaction applied from 2 August 2026 regardless.

Sometimes, and the boundary is finer than the reassuring version allowed. The deployer duty for AI-generated text sits in Article 50(4), second subparagraph, and covers text "published with the purpose of informing the public on matters of public interest". The Commission's FAQ enumerates that, and the list ends with an open clause: "any economic, financial, political, scientific, or cultural developments that may be relevant subject of public debate." A good deal of B2B thought leadership lives inside it.

The Guidelines settle the marketing question with worked examples. Outside the duty: "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)." Read the parenthetical carefully, because it pulls a product page carrying a sustainability or safety claim back inside. Listed inside the duty is an AI-manipulated lifestyle-website article on the effects of diets on a disease, which is a commercial publisher's content piece. No regulator has drawn this line in a decision yet, so treat it as a planning assumption rather than a settled boundary.

A named human who reviews the work and carries responsibility for publishing it. Article 50(4) lifts the disclosure obligation where two cumulative conditions are met: the text has undergone human review or editorial control, and a legal or natural person holds editorial responsibility for the publication. That exemption does not depend on winning the scope argument, which is what makes it the reliable control.

The bar sits well above a proofread. The Commission defines human review as "the deliberate examination of the substance of the content by one or more natural persons possessing relevant knowledge and professional judgement pertaining to the subject matter under scrutiny (e.g. academic peer review or professional validation chains)", and adds that "fact-checking the accuracy of the content is a minimum requirement". It rules out the shortcut directly: "Superficial, solely formal or procedural checks (e.g. spell-checking or grammatical correction), the mere existence of an editorial policy, automated review processes or cursory editorial approval without substantive engagement" cannot satisfy the condition.

One ordering rule decides whether any of it works, and it is the likeliest way a team loses the exemption without noticing. Where AI systems modify, supplement or reformulate content after editorial sign-off, "any substantive AI intervention occurring after the human review or editorial control process has taken place will therefore cause the exception to become void". An AI tool that rewrites your meta description, reformats the piece for social or translates it after approval undoes the protection you just built. The human has to be last.

Two limits finish the picture. Editorial responsibility means holding "the ultimate legal responsibility over the publication of the content, including the human review or editorial control", and that person's identity and contact details "should be made publicly available on an easily findable location", such as a website's terms. And the exemption is written for text alone. It does nothing for the image, audio and video duty in the first limb of the same article, nothing for the obligation to tell someone they are speaking to an AI, and nothing for machine-readable marking.

Only the ones that are deep fakes. The first limb of Article 50(4) is a separate duty on deployers covering image, audio and video, and it bites on deep fakes, defined in Article 3(60) 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". Synthetic presenters and synthetic influencers testing a real product are squarely inside. Colour correction, background replacement and re-scaling in a product ad generally are not, and nor is obviously fantastical imagery.

There is no artistic carve-out, whatever you may have read. The Commission states that deep fakes forming part of evidently artistic, creative, satirical or fictional works "are not excluded from the transparency obligation", and that deployers "still need to disclose the AI-origin of the content or its manipulation" in a manner that does not hamper enjoyment of the work. What exists is a lighter way of disclosing rather than permission to skip it, and where a deep fake mixes characters, "the informative character should always prevail".

Yes, and the duty is written onto the provider rather than onto whoever uses the tool. Article 50(1) requires AI systems intended to interact directly with natural persons to be designed so those people are informed they are interacting with AI, unless that is obvious "from the point of view of a natural person who is reasonably well-informed, observant and circumspect, taking into account the circumstances and the context of use". Article 50(5) fixes the timing at the latest at the time of the first interaction or exposure.

Whether the duty is yours turns on a definition, not on who owns the website. Article 3(3) makes you a provider where you develop an AI system, or have one developed, and put it into service under your own name or trademark. License a chat widget and run it as the vendor's product, and the design obligation is the vendor's. Build your own booking qualifier, or badge a white-label agent as yours, and you are its provider. The Article 50(2) marking duty lands in the same place, on the provider of the generative system.

Mostly an inventory and one ordering change, not a build. Article 99(4) sets fines for breaching Article 50 at up to 15 million euro or 3% of total worldwide annual turnover, whichever is higher, with Article 99(6) capping SMEs and start-ups at the lower of the two instead. Compliance "will mainly be enforced by national competent market surveillance authorities", so intensity will vary between member states.

• List every place an AI system speaks directly to a person, then establish for each whether you are the provider under Article 3(3) or a deployer of someone else's product
• Move human review to the end of the workflow, after every AI step, because a substantive AI intervention after sign-off voids the exemption
• Make the review substantive and fact-checked by someone who knows the subject, since spell-checking and cursory approval are expressly ruled out
• Publish the identity and contact details of whoever holds editorial responsibility, in your terms or other user-facing legal information
• Check any product page carrying a health, consumer-safety or sustainability claim, because that is the carve-back that pulls commercial copy into scope
• Audit paid creative for synthetic presenters or manipulated video, a deployer duty your tooling vendor does not carry for you

At the end, after every automated step. That is how we sequence the engines we build, and we would sequence them that way regardless of Article 50, because a machine that writes without a competent person closing the loop produces work nobody will stand behind. Whether that sequencing discharges your own obligations under the Regulation is a question for your counsel.