EU AI Act for Advertisers: What to Disclose Starting August 2026

If you run ads that reach EU audiences or sell to New York consumers, you have two new compliance dates on the horizon. EU AI Act Article 50 kicks in on August 2, 2026. The New York AI Transparency in Advertising Act (Bill S.8420-A / A.8887-B) becomes effective June 9, 2026.

Most legal-blog coverage I’ve read takes a “label everything AI” panic approach. After working through the actual statutory texts and the analyses from Jones Day, Skadden, Kelley Drye, and Bird & Bird, I think that framing is wrong. Both laws target a much narrower scope than the headlines suggest.

A note before we start. This is not legal advice. If you need a binding compliance opinion for your specific situation, talk to a qualified lawyer in advertising or IT law. What follows is my read as an advertiser, not as counsel.

---

The headline you actually need

Both laws target AI-generated human likenesses, not AI-generated images in general. A render of a product, a photo of a dog, an abstract illustration, a stylized graphic. None of these trigger disclosure under either law.

What does trigger disclosure. Realistic AI-generated humans, deepfakes of real people, lifestyle imagery where the AI person looks authentic enough to deceive a viewer.

If your brand uses product photography and stock images of real licensed people, you have nothing to do. If you use AI-generated lifestyle scenes with synthetic humans, you need a disclosure plan before the deadlines hit.

---

The two laws are not the same

Stop conflating them. EU AI Act and NY Synthetic Performer Act work differently.

EU AI Act Article 50

In force August 2, 2026. The deepfake definition lives in Article 3, paragraph 60. AI-generated or manipulated content that resembles existing persons, objects, places, entities or events and would falsely appear to be authentic.

Three criteria, all required.

1. Artificially generated or manipulated
2. Resembles an existing entity (person, place, etc.)
3. Would appear authentic to a viewer

The exceptions are explicit and useful. Article 50 paragraph 4 carves out artistic, creative, satirical, fictional or analogous work. Where transparency obligations are limited to disclosure of the existence of generated content in a manner that does not hamper the display or enjoyment of the work.

For text content. AI-generated text on matters of public interest must be labeled, unless it has been subject to human review or editorial control with a natural or legal person bearing editorial responsibility.

Penalties. Up to 15 million euros or 3 percent of worldwide annual turnover, whichever is higher. Not symbolic.

New York AI Transparency in Advertising Act (S.8420-A)

In force June 9, 2026. The Act introduces “synthetic performers” as a defined category. AI-generated assets intended to look like human performers in advertising.

Triggers, all three required.

1. AI-generated synthetic performer (human likeness)
2. Used in commercial advertisement
3. Reaches a New York audience

Penalties. $1,000 first violation, $5,000 each subsequent. Per advertisement, not per impression. But the math compounds quickly across a multi-asset campaign.

Exceptions. Expressive works including motion pictures, TV programs, streaming content, documentaries, video games, audiovisual works. The exception applies if the synthetic performer’s use is consistent with the expressive work itself.

The key difference

EU AI Act covers all deepfakes regardless of medium. NY targets specifically synthetic performers in commercial advertising. EU is broader on content type, NY is more specific on context.

If you advertise in EU and NY, you need both compliance frameworks. They overlap but don’t substitute.

---

Cross-jurisdictional triage matrix

Apply this to every image and video in your live campaigns and landing pages.

Image Type	EU AI Act	NY Synthetic Performer
Pure product render, no person	🟢 Safe	🟢 Safe
Animal photography (dog, cat, etc.)	🟢 Safe	🟢 Safe
Atmospheric or setting shot, no person	🟢 Safe	🟢 Safe
Stylized AI illustration, comic, abstract	🟢 Safe (Article 50(4) creative exception)	🟢 Safe
AI-modified real photo (color, touch-up)	🟡 Check, depends on “appears authentic” test	🟡 Check, depends on human likeness alteration
Realistic AI photo of fictional person	🔴 Required (deepfake definition)	🔴 Required (synthetic performer)
AI lifestyle image with AI human	🔴 Required	🔴 Required
Deepfake of real person without consent	🔴 Required + GDPR + right of publicity	🔴 Required + posthumous publicity rights for deceased

The 🟡 row is where you need a lawyer. Specific facts matter. The 🟢 and 🔴 rows are clear enough that you can act yourself.

---

Beyond EU and NY. The federal patchwork

US state-level AI advertising laws are accelerating. California AI Transparency Act (effective August 2, 2026) requires platforms like LinkedIn to provide AI detection tools and watermark support. Colorado, Utah, and Minnesota have related laws either passed or in legislative pipeline. The FTC has signaled aggressive enforcement against deceptive AI content under existing consumer protection statutes.

The compliance overlap with tracking obligations is real. If your AI disclosure metadata flows through Google Ads creative auto-generation or Meta’s Advantage+ asset enhancements, the same data infrastructure your server-side tracking setup handles is what carries the disclosure flags. Build them together, not as separate projects.

In Europe, beyond the AI Act itself, member states are activating UWG-style unfair competition statutes. In Germany, Wettbewerbszentrale (the Centre for Protection against Unfair Competition) and competitor brands can issue warning letters under § 3a UWG. Similar mechanisms exist in Austria and Switzerland.

The cross-jurisdictional reality. If you’re a brand selling globally, you’re now navigating a compliance patchwork. Standardize on the strictest applicable framework and you cover most of the rest.

---

What to do this week

Practical, 30-minute audit per brand.

1. List every live ad image, video, and landing page hero image
2. Categorize each. Custom shot, licensed stock, AI-generated without person, AI-generated with person, AI-modified real photo
3. Apply triage matrix. 🟢 / 🟡 / 🔴
4. For 🔴 items. Plan disclosure (textual label, watermark, caption) by July 2026
5. For 🟡 items. Get legal opinion or default to disclosure
6. For 🟢 items. Document the determination so you can defend it if challenged

The cost of this audit is one afternoon. The cost of skipping it. Potential fines up to 15M EUR (EU) or $5,000 per violation (NY) plus competitor lawsuits in DACH markets.

---

Disclosure formats that work

The EU Code of Practice on AI-Generated Content (final version expected June 2026) will provide more concrete standards. Until then, these formats are recognized in current legal practice.

Textual labels.

• “AI-generated”
• “Created with AI”
• “AI image”

Visual marks.

• “AI” or “Generated by AI” watermark embedded in image
• Pictogram in corner of image, visible without hover

Captions.

• Below or adjacent to image, with clear visual association
• Not in alt-text alone, not in footer disclaimers

The principle. The disclosure must be where the image is, not buried elsewhere. Hidden or hard-to-find disclosure does not satisfy either law.

---

Brand example walkthroughs

E-commerce fashion brand. You use AI-generated models to showcase product. 🔴 Required. Both laws trigger. Solution. Either watermark every image with “AI model” tag, or switch to real model photography for hero campaigns and use AI only for catalog grid where viewers can clearly tell from context. Most brands will need a disclosure system.

SaaS B2B brand. You use stock photos of “developers at laptops” that turn out to be AI-generated. 🔴 Required if AI-generated and reaching EU audience. Solution. Audit your stock library. Many premium stock providers now flag AI-generated content. According to a 2025 industry survey, around 30% of “premium” stock photo libraries now contain AI-generated assets without clear labeling at the metadata level. Replace where needed or add disclosure. The same audit reveals gaps that often cost you in Google Ads conversion tracking accuracy.

Local service business with website ads. You use real photos of your actual team and location. 🟢 Safe. Solution. Document that you use real photography. No further action.

Healthcare or pharmaceutical brand. HIPAA / GDPR / sectoral regulations layer on top. AI-generated patient images would be 🔴 plus additional regulatory issues. Consult sector-specific counsel.

---

Sources I read

I link only original statutory text and high-quality law-firm analyses. Skipping marketing blogs.

EU AI Act official.

• Article 50 official text on the EU Commission AI Act Service Desk
• Annotated version on artificialintelligenceact.eu
• Code of Practice on AI-Generated Content draft, EU Commission Shaping Europe’s Digital Future portal

NY Synthetic Performer Act.

• NY Senate Bill S.8420-A and Assembly Bill A.8887-B official texts on nysenate.gov
• Governor Hochul’s signing statement on governor.ny.gov

Law firm analyses (high quality).

• Jones Day on the Code of Practice draft
• Skadden on the two new NY laws
• Kelley Drye on synthetic performer disclosure
• Davis+Gilbert on AI legal updates and federal-state conflict
• Hunton on the NY Act
• Akerman on brand and franchise compliance reshaping
• Bird & Bird on the transparency code

All URLs are in my sources document on GitHub. Search for “eu-ai-act-2026-sources”.

---

FAQ

Does the EU AI Act apply to small advertisers?

Yes. Article 50 applies to deployers of AI systems regardless of company size when ads reach EU audiences. The threshold is the content type (deepfake / human likeness), not company revenue. Small brands using AI-generated lifestyle imagery have the same disclosure obligation as enterprises.

What if my AI image only modifies a real photo lightly?

This is the 🟡 zone. The EU test is whether the modified image “would appear authentic to a viewer.” A color correction or background replacement on a real photo of a real person typically does not trigger Article 50. A face-swap or generative re-imagining does. When in doubt, document the determination and consult counsel.

Do I need to disclose AI-written ad copy?

Article 50 covers AI-generated text only when published “to inform the public on matters of public interest” without human review. Standard sales copy, product descriptions and ad headlines are typically out of scope, especially when reviewed by a human marketer before publish. Editorial articles, news, and political advertising are different categories with stricter rules.

How does the NY Act interact with German UWG abuse warnings?

They are independent. NY Act gives state attorney general enforcement power. German § 3a UWG gives competitors and Wettbewerbszentrale the right to issue cease-and-desist letters. A non-compliant ad campaign running in DACH and reaching New York consumers can face action under both regimes simultaneously.

What about AI-generated voiceovers in video ads?

Both laws extend to audio. EU Article 50 covers AI-generated audio of identifiable persons. NY Synthetic Performer Act covers synthetic voice performers. A clearly synthetic narrator voice (TTS-style) is generally lower-risk than a voice cloned from a real person.

---

Next steps for your account

If your campaigns run in DACH or New York and use any AI imagery with humans, the 30-minute audit above is overdue. I do compliance audits as part of larger Google Ads engagements, including how disclosure interacts with Performance Max creative auto-generation and Smart Bidding signal quality. Account-level negative keyword hygiene from the negative keywords workflow often surfaces compliance issues alongside the obvious budget waste.

In one audit, 64% of the brand’s Performance Max creative library was AI-generated lifestyle imagery, none of it labeled. The remediation path is straightforward once the inventory exists.

Schedule a free intro call to discuss your specific exposure and a defensible disclosure approach.