EU AI Act Article 50 is not a blanket rule that every AI ad needs a sticker. From 2 August 2026 it splits duties: providers must make synthetic outputs machine-detectable, while deployers who publish deepfake-class image, audio, or video must disclose artificial generation or manipulation in a clear, human-perceivable way at first exposure. Brands and agencies care because advertising is a deployer context — and machine metadata alone is not enough when disclosure applies.
This is a practical guide for EU and UK brand and agency teams working with AI film studios. It is not legal advice. Final labelling decisions belong with your counsel. AIQ Studio, based in Belgrade and serving EU/UK buyers, prepares disclosure-ready masters and delivery notes; we do not sell “AI Act certified” guarantees.
What Article 50 covers (high level)
Article 50 of Regulation (EU) 2024/1689 sets transparency obligations for certain AI systems. The European Commission’s FAQ on transparency obligations under Article 50 is the clearest public entry point for non-lawyers. The statutory text is also published on the AI Act Service Desk.
At a useful altitude for advertisers, four ideas matter:
- Providers design and place AI systems on the market; they carry machine-readable marking duties for many generative outputs (Article 50(2)).
- Deployers use AI systems under their authority in a professional capacity; advertising companies typically sit here when they publish ads.
- Deepfake-class image, audio, or video triggers deployer disclosure that must be clear to people, not only to machines (Article 50(4) and 50(5)).
- Separate rules cover direct AI interaction (chatbots), emotion recognition / biometric categorisation notices, and some public-interest text — most routine ad copy sits outside the text limb unless it informs the public on matters of public interest without editorial control.
Recommendation: map your campaign once: who is the provider of the generative system, who is the deployer publishing the ad, and which limb of Article 50 might apply. Do not collapse everything into “we used AI, so label it.”
Why advertisers care from August 2026
Article 50 applies from 2 August 2026. The Commission FAQ states that providers and deployers must comply from that date. A limited grace period to 2 December 2026 is described for certain Article 50(2) marking obligations for systems already placed on the market before 2 August 2026 — it is not a general holiday from deployer deepfake disclosure.
Content generated before 2 August 2026 generally does not need retroactive labelling under the FAQ’s framing, though counsel may still prefer voluntary labels on long-running assets. Enforcement sits mainly with national market surveillance authorities. The FAQ notes fines can reach up to €15 million or 3% of total worldwide turnover for the preceding financial year, with proportionality considerations for smaller companies.
Recommendation: treat August 2026 as an operational deadline for new creative pipelines, not a theoretical compliance seminar. Build labelling options into delivery specs now if your ads will still run after that date.
Deepfake-class video vs ordinary CGI — practical framing
Under the Act’s definition (Article 3(60), explained in the Commission FAQ), a deepfake is AI-generated or manipulated image, audio, or video that resembles existing persons, objects, places, entities, or events and would falsely appear authentic or truthful. Three cumulative ideas keep showing up in Commission materials: resemblance, an existing or plausibly existing referent, and a false appearance of authenticity.
That is broader than “fake celebrity scandal clips.” Photoreal AI ads that present invented scenes as if they were filmed truths can fall into the analysis. At the same time, Commission guidance materials discuss contexts where audiences do not expect authenticity — for example certain artistic or standard film-production effects — and Article 50(4) limits disclosure for evidently artistic, creative, satirical, fictional, or analogous works to an appropriate manner that does not hamper enjoyment.
Ordinary CGI and assistive editing also matter on the provider side: Article 50(2) marking does not apply to the same extent where systems perform assistive standard editing without substantially altering input semantics. That is a technical legal line for providers and tools — not a free pass for every glossy ad.
Recommendation: ask counsel for a campaign-level deepfake assessment using the resemblance + authenticity test in your deployment context. Studios can describe process; they should not self-certify the legal outcome.
Human-perceivable disclosure vs machine marks
Two different transparency tools get mixed up in briefings.
| Mechanism | Who mainly owns it | What it does | What it does not do |
|---|---|---|---|
| Machine-readable marks | Providers (Art. 50(2)) | Helps detection that content is synthetic or manipulated | Does not replace human-facing deployer disclosure |
| Human-perceivable disclosure | Deployers (Art. 50(4)/(5)) | Tells people, clearly, at first exposure | Is not satisfied by invisible metadata alone |
The Commission FAQ is explicit: deployers cannot simply rely on provider machine-readable marking to fulfil deepfake disclosure. Disclosure must be clear, distinguishable, understandable, and perceivable without special tools — for example visible or audible labels — at the latest at first exposure, and it must meet accessibility requirements.
Recommendation: if disclosure applies, budget for on-creative or immediately adjacent labels across every cutdown and platform crop. A label that dies in a 9:16 safe zone is not a plan.
What studios can prepare (labels, delivery notes)
A film studio cannot replace your lawyer. It can make compliance work less chaotic.
Disclosure-ready delivery stack:
Locked master → clean and labelled variants → delivery note (AI methods used, formats, recommended label placement) → client legal sign-off → publish
- Deliver clean masters without a forced label when counsel wants to decide placement.
- Deliver labelled variants with a clear, high-contrast disclosure treatment coordinated with brand guidelines.
- Provide a short delivery note: generative methods used at a useful altitude, formats delivered, and where labels sit in 16:9 and 9:16.
- Keep Spec honesty separate: portfolio frames not commissioned by a brand stay labelled as Spec. Not commissioned by the brand shown.
- Refuse fake certificates. “Disclosure-ready” is a delivery posture, not a conformity mark.
How we talk about process and craft is on Studio. Commercial scopes and payment terms are on Pricing.
Recommendation: put “labelled / unlabelled masters + delivery note” in the statement of work before production starts. Do not invent labels in the last export hour.
What only client legal can decide
Only your legal and compliance teams should decide whether a specific ad is a deepfake under Article 50, whether artistic exceptions change the disclosure form, whether public-interest text rules capture any accompanying copy, and how national enforcement practice plus other regimes (IP, personality rights, advertising codes, consumer law) interact.
UK brands distributing into the Union still need EU-facing analysis when ads are shown in the EU. A production vendor outside the EU does not erase deployer duties for the legal person publishing under their authority.
Recommendation: keep studio notes factual (“methods used”, “label variants delivered”) and keep legal conclusions in counsel memos. Mixing the two creates false comfort.
How AIQ phrases disclosure-ready delivery
AIQ Studio is a registered company in Belgrade (legal IDs to be added on site when ready). We produce AI commercials and branded films for EU/UK clients with fixed packages and watermarked previews before payment. On Article 50 we use restrained language:
- We support disclosure-ready delivery: clean masters, optional labelled variants, and a delivery note.
- Final labelling is coordinated with the client’s legal team.
- We do not claim AI Act certification, guaranteed compliance, or that every AI commercial requires an on-screen label.
Contractual framing also lives in Terms. If you need a production partner that treats transparency as part of craft rather than a footnote, send the brief through Contact.
Sources
- European Commission FAQ — Transparency obligations under Article 50 of the AI Act
- AI Act Service Desk — Article 50
- Commission materials on guidelines for AI transparency obligations
Law-firm explainers can help orientation, but primary sources above govern how we phrase public guidance.
What to do next
- Ask counsel for a one-page Article 50 map per campaign (provider vs deployer; deepfake risk; label form).
- Add labelled/unlabelled master requirements to your production brief.
- Align platform crops so disclosure remains perceivable on 9:16.
- Brief an AI film studio that will document methods without overclaiming compliance.
Ready to produce with disclosure options in the delivery pack? Start a project.
FAQ
Do all AI ads need an on-screen label?
No. Article 50 does not automatically require an on-screen label on every AI-assisted advertisement. Deployer disclosure focuses on deepfake-class image, audio, or video (and separate text rules for certain public-interest publications). Whether your ad meets the deepfake test is a legal assessment.
Is this legal advice?
No. This article is general information for brand and agency teams. It is not legal advice. Use Commission sources and your counsel for decisions that affect publishing risk.
Does a Serbia-based studio change EU deployer duties?
Not by itself. Deployer duties turn on who uses the AI system under their authority and where content is exposed. A Belgrade production vendor can support delivery; the brand or agency publishing in the EU still needs its own compliance analysis. This is not legal advice.
Are machine-readable AI marks enough?
No for deployer deepfake disclosure. The Commission FAQ states deployers cannot rely only on provider machine-readable marks; disclosure must be human-perceivable at first exposure when Article 50(4) applies.
When does Article 50 start to apply?
From 2 August 2026, according to the Commission FAQ. A limited later date (2 December 2026) is discussed for certain provider marking duties on systems already on the market before 2 August 2026 — not as a general delay for deployer deepfake disclosure.
What can AIQ deliver for compliance teams?
Clean masters, optional labelled variants, and a delivery note describing methods and formats at a practical level — coordinated with your legal team. See Studio and Pricing for process and packages.