EU AI Act Article 50 for Marketers: What to Disclose, and the December 2026 Deadline
Article 50 of the EU AI Act sets transparency obligations for AI systems that interact with people or generate content. It has applied since 2 August 2026. If your team publishes AI-written or AI-generated marketing material into the EU, it reaches you — not only the vendor who built the model.
This is a practical read for marketing teams, not legal advice. Every obligation below links to a primary source, and you should read those before making a compliance decision.
The four obligations that touch marketing work
1 · Synthetic content must be machine-readably marked. Providers of AI systems that generate synthetic audio, image, video or text must ensure outputs are marked in a machine-readable format as artificially generated or manipulated. This duty sits with the provider of the generating system — which means it is a question for your vendor, not a box you tick yourself.
2 · Deepfakes must be disclosed. Deployers of an AI system that generates or manipulates image, audio or video constituting a deepfake must disclose that the content is artificially generated or manipulated. The deployer is the organisation publishing it. That is you.
3 · Text on matters of public interest must be disclosed. Where AI-generated or manipulated text is published to inform the public on matters of public interest, deployers must disclose that it was artificially generated. Most product marketing sits outside this; thought leadership on public policy, health, or safety topics may not.
4 · People must be told when they are talking to an AI, unless it is obvious from the context. Relevant to any chat widget or voice agent on your site.
The full text is at the AI Act Service Desk, and the European Commission maintains an FAQ on Article 50 plus guidelines on transparency obligations.
The dated item: 2 December 2026
Under the AI Omnibus provisional agreement of May 2026, generative AI systems already on the market before 2 August 2026 have until 2 December 2026 to meet the machine-readable marking requirement in Article 50(2).
The practical read: if you operate in the EU and your stack includes a generative tool you were using before August, that grace period is running out. It is worth putting one question to that vendor in writing now — does your output carry machine-readable marking under Article 50(2), and from what date? — and keeping the answer.
You want the answer in writing for the same reason you keep anything in writing: the duty to mark sits with the provider, but the consequences of publishing unmarked synthetic content land on the brand whose name is on it.
How this differs from the FTC's rules
Teams operating on both sides of the Atlantic tend to collapse these into one "AI disclosure" policy. They are different in kind, and the difference matters.
The EU rule is about transparency. Disclose that content is synthetic, and the content is permitted.
The FTC rule on fake and AI-generated reviews and testimonials is a prohibition. Finalised in August 2024 and enforceable since October 2024, it bans testimonials from people who do not exist. Disclosure does not cure it — the problem is not that the reader was misled about the origin, it is that no real customer holds that opinion. Labelling a fabricated testimonial as AI-generated does not make it lawful; it makes it a labelled fabrication.
Liability there is not limited to the brand, either: agencies and intermediaries who produce or place the content are within reach of the rule. If you run marketing for clients, this is your exposure too, not just theirs.
Existing endorsement and substantiation rules apply unchanged to AI-written copy. See the FTC's endorsement guidance.
The distinction most policies get wrong
"AI-generated" and "AI-assisted" are not the same disclosure, and collapsing them helps nobody.
A human-written piece checked by a model is not synthetic content. A model-written piece lightly edited by a human is. Most teams sit somewhere on that spectrum for most of their output, and very few have decided where their line falls.
Decide where yours is. Write it into your brand rules. Then apply it the same way every time — consistency is what a regulator, a client, or a journalist will actually look for, far more than the specific threshold you picked.
A practical checklist
Three habits cover most of the exposure on both sides:
- Keep a record of what was generated, checked and approved — by whom, and when. Append-only, exportable. This is the single artefact that answers almost every question either regime raises. (What an audit trail should record.)
- Never let a synthetic person endorse you. No fabricated testimonials, no AI-generated "customers", no invented case studies. This is a bright line, not a judgment call.
- Substantiate every factual claim before it ships, exactly as you would for a human-written ad. A claim with no evidence behind it should not be available to the agent in the first place — which is what a claims list with sources is for.
Then four specifics:
- Ask your generative vendors, in writing, about Article 50(2) marking and their date.
- Label any chat or voice agent on your site as AI, unless it is unmistakable from context.
- Write your AI-generated / AI-assisted line into your brand rules, with a date.
- Check whether any of your content informs the public on matters of public interest — and if it does, disclose.
Teams that already run a real approval workflow with a recorded sign-off have very little work to do here. Teams that do not will find that the compliance requirement and the quality requirement turn out to be the same requirement, which is the more useful way to think about all of it.
Azimuth records every generation, flag, approval and change as an append-only trail that exports to CSV, which is the artefact most of the above comes down to. We collected the dated obligations, a four-part grammar for compliance flags, an eight-point pre-publish check and a brand-rules template into a free field guide — no invented statistics, every regulatory claim linked to its primary source. Download The Agentic Marketing Field Guide.
This post is a summary of public regulatory material and is not legal advice. Read the primary sources linked above before making a compliance decision.