The three deepfake tests (all must be true)
The Commission’s Article 50 FAQ (last update 24 July 2026) cites Article 3(60) of the AI Act and lists three cumulative criteria:
| Test | What it means for marketing |
|---|---|
| Resemblance | High similarity to the simulated subject (a real CEO face, a known product, a real storefront). |
| Existing | The subject exists, can plausibly exist, or could have existed in reality. |
| False authenticity | A person could take the clip as real. Context and audience expectations matter. |
A fully synthetic product demo that looks like a real customer interview usually fails the authenticity test in your favour only if the audience cannot reasonably treat it as real. Treat that as a narrow call. Most founder UGC-style AI ads fail the other way.
What the Commission said on 31 July
From 2 August 2026 the Commission’s AI Office and national authorities enforce the AI Act. On the same date the transparency rules start. The 31 July 2026 press release is plain: deepfakes (images, videos, or audio edited or generated with AI) must be labelled. AI-generated or altered content must also carry machine-readable marks so it can be detected.
That second duty is mostly on providers under Article 50(2). Your deployer duty under Article 50(4) is the human-visible label. The FAQ is explicit: you cannot rely only on the provider’s machine-readable mark.
Marketing cases that usually need a label
- AI face-swap or voice-clone of a real customer, founder, or influencer in an ad.
- AI video that looks like a real interview, testimonial, or news clip about your product.
- AI audio of a real person’s voice reading your script for paid social or IVR.
- AI-edited photos of a real office, store, or person that would pass as documentary.
Personal, non-professional deepfakes sit outside the Act. Paid ads, agency work, and freelancers acting for your company do not. The FAQ treats economic activity as professional deployer use.
What usually does not need a 50(4) deepfake label
- Obviously cartoon, stylised, or fictional brand mascots where authenticity is not claimed.
- Evidently artistic, creative, satirical, or fictional works (limited disclosure that does not spoil the work).
- Standard editing assist (light cleanup) that stays inside the provider marking exemptions in the guidelines.
- AI text on your blog that went through real human review with editorial responsibility (that is the text limb of 50(4), not deepfakes).
EU icons for AI-generated content are optional helpers. Labelling itself is not optional. Icons alone do not prove compliance.
What to do this week
- Inventory every AI image, video, and audio asset currently live in EU-facing ads, landing pages, email, and social.
- Flag anything that resembles a real person, place, or event and could pass as authentic.
- Add a clear, distinguishable label at first exposure (on-frame text or audible cue). Do not bury it in Terms.
- Keep provider machine-readable marks where the tool supports them. Open a 2 December 2026 marking ticket only for generative systems already on the market before 2 August.
- Do not open a high-risk Annex III programme because of this page. That calendar is December 2027.
Need a print-and-tick inventory that covers deepfake labels, chatbot notice, and the December marking ticket? That is the $29 checklist. Need a memo on whether your specific ad set is in or out? That is the $149 Money Brief.
Buy the operator pack
The $29 Article 50 Ship Checklist is the inventory, notice line, deepfake labels, and 2 December ticket. The $149 Money Brief is the sourced memo if your creative is borderline, you use a celebrity likeness, or a national authority has written.
Free notice-copy tool · Does Article 50 apply to my chatbot?
Sources
Fetched 28 August 2026 (PT). No invented quotes.
- European Commission, Article 50 FAQ, last update 24 July 2026 (deepfake definition Art. 3(60); three cumulative criteria; deployer label on first exposure; cannot rely only on Art. 50(2) marks). Transparency obligations under Article 50 of the AI Act
- European Commission, “Commission starts enforcing AI Act rules and new transparency requirements on 2 August,” press release, 31 July 2026. digital-strategy.ec.europa.eu
- European Commission, EU Icons for labelling AI-generated content, last update 10 August 2026 (icons optional; labelling mandatory). digital-strategy.ec.europa.eu/en/policies/eu-icons-labelling-ai-generated-content
- European Commission, “Commission publishes guidelines on transparency obligations,” 20 July 2026. digital-strategy.ec.europa.eu