Resources · Cheat sheets · October 2026 · v1.0

EU AI Act Disclosure Cheat Sheet

Q: What does Article 50 of the EU AI Act require for AI-generated content?

Article 50 of the EU AI Act creates four transparency duties: (1) tell people when they are interacting with an AI system, at the start; (2) providers must embed effective machine-readable marks on synthetic text, image, audio and video and supply detection means; (3) deployers must label deepfakes and AI-generated text published on public-interest matters, unless it had genuine human review with editorial responsibility taken; (4) disclose when emotion-recognition or biometric categorization systems are running. Every disclosure must be clear, distinguishable and accessible at first exposure.

Q: When did the EU AI Act transparency obligations take effect?

Article 50 transparency obligations began applying on 2 August 2026. Prohibited AI practices were banned from 2 February 2025 and general-purpose AI model obligations applied from 2 August 2025. A grace period runs to 2 December 2026 for marking on AI systems placed on the market before 2 August 2026.

Q: What are the fines for EU AI Act transparency violations?

Under Article 99 the EU AI Act has three fine tiers: up to €35 million or 7% of global annual turnover for prohibited practices; up to €15 million or 3% for transparency and other obligations, which includes Article 50 disclosure duties; and up to €7.5 million or 1% for supplying incorrect or misleading information to authorities. For large companies the higher of the fixed amount or the percentage applies; SMEs and startups get proportionate caps.

Q: What is the difference between an AI Act provider and a deployer?

A provider builds or sells the AI system, such as model vendors and AI tool makers, and owes machine-readable marking plus detection means. A deployer uses AI and publishes AI-generated content, such as marketers, publishers and agencies, and owes labeling of deepfakes and AI text plus disclosure of AI interaction. One company can be both.

By the Ingenious AI team

Every EU AI Act disclosure requirement for AI-generated content, on one page: the Article 50 transparency duties, provider vs deployer roles, key dates, fine tiers, the AI-generated content labeling rules on YouTube, Meta, TikTok and Google Ads — and a 60-second self-check. Download the free one-page PDF, ungated.

⬇ Download the one-page PDF (free, no signup)

1. The 4 transparency duties — Article 50

Disclose AI interaction. Chatbots, voice agents and AI assistants must tell people they are talking to AI — at the start of the interaction.
Mark synthetic media. Providers must embed effective machine-readable marks (watermark / C2PA) on AI-generated text, image, audio and video — and supply detection means.
Label what you publish. Deployers must label deepfakes (realistic synthetic people, voices, events) — and AI-generated text published on public-interest matters, unless it had genuine human review with someone taking editorial responsibility.
Disclose emotion & biometric systems. Tell people when emotion-recognition or biometric categorization is running on them.

Rule of thumb: every disclosure must be clear, distinguishable and accessible — at first exposure. Never buried in terms and conditions or hidden in menus.

2. Provider vs deployer — who owes what

Provider — builds or sells the AI system (model vendors, AI tool makers). Duty: machine-readable marking + detection means on synthetic outputs.
Deployer — uses AI and publishes AI content (marketers, publishers, agencies). Duty: label deepfakes and AI text, disclose AI interaction.

One company can be both — for example, an agency using a vendor model in client campaigns.

3. Key dates

2 February 2025 — Prohibited AI practices banned (Article 5).
2 August 2025 — General-purpose AI (GPAI) model obligations apply.
2 August 2026 — Article 50 transparency duties apply — in force now.
2 December 2026 — Grace period ends: AI systems placed on the market before August 2026 must carry marking.

4. Fine tiers — Article 99

€35M or 7% — Prohibited practices (Art. 5): the highest tier in EU digital regulation.
€15M or 3% — Transparency and other obligations, including all Article 50 duties on this sheet.
€7.5M or 1% — Incorrect or misleading information supplied to authorities.

Whichever is higher — the fixed amount or the percentage of global annual turnover — for large companies; proportionate caps apply for SMEs and startups.

5. Platform label rules

YouTube — "Altered or synthetic content": disclose realistic AI-made or AI-altered video at upload. Mandatory since May 2025.
Meta (Facebook · Instagram · Threads) — "AI info" / "Made with AI" labels applied automatically via C2PA metadata and detection. Undisclosed AI in ads can mean ad rejection.
TikTok — "AI-generated" label required on realistic AI images, audio and video. Set the toggle at upload.
Google Ads — Realistic AI depictions in ads must be disclosed ("Generated by AI" labeling).

One clip posted to three platforms must be evaluated and labeled in each upload flow. Platform rules sit alongside the EU AI Act — you must satisfy both.

6. The 60-second self-check

Can a viewer tell at first glance that each AI-generated asset is AI-made?
Are deepfakes — realistic synthetic people, voices, events — clearly labeled?
Is AI-written public-interest text labeled — or genuinely human-reviewed with editorial responsibility?
Do your chatbots and voice agents disclose they are AI at the start?
Do your tools embed machine-readable provenance (C2PA / watermark)?
Can you prove it — a per-asset audit trail if a regulator asks?

A "no" anywhere is a gap. This sheet teaches the rules once — continuous monitoring closes them forever.

From checklist to always-on

DisclosureGuard reads declared and embedded provenance — C2PA, watermarks, metadata — never guesswork, and scores every asset against the rules above. This cheat sheet covers the rules once; DisclosureGuard checks them on every asset, 24/7.

Check every asset free →

Frequently asked questions

What does Article 50 of the EU AI Act require for AI-generated content?

Article 50 creates four transparency duties: (1) tell people when they are interacting with an AI system, at the start; (2) providers must embed effective machine-readable marks on synthetic text, image, audio and video and supply detection means; (3) deployers must label deepfakes and AI-generated text published on public-interest matters, unless it had genuine human review with editorial responsibility taken; (4) disclose when emotion-recognition or biometric categorization systems are running. Every disclosure must be clear, distinguishable and accessible at first exposure.

When did the EU AI Act transparency obligations take effect?

Article 50 transparency obligations began applying on 2 August 2026. Prohibited AI practices were banned from 2 February 2025 and general-purpose AI model obligations applied from 2 August 2025. A grace period runs to 2 December 2026 for marking on AI systems placed on the market before 2 August 2026.

What are the fines for EU AI Act transparency violations?

Under Article 99 there are three fine tiers: up to €35 million or 7% of global annual turnover for prohibited practices; up to €15 million or 3% for transparency and other obligations, which includes Article 50 disclosure duties; and up to €7.5 million or 1% for supplying incorrect or misleading information to authorities. For large companies the higher of the fixed amount or the percentage applies; SMEs and startups get proportionate caps.

Do YouTube, Meta, TikTok and Google Ads require AI disclosure labels?

Yes. YouTube requires creators to disclose realistic altered or synthetic content at upload (mandatory since May 2025). Meta applies "AI info" / "Made with AI" labels automatically via C2PA metadata and detection, and undisclosed AI in ads can trigger ad rejection. TikTok requires the "AI-generated" label toggle at upload for realistic AI content. Google Ads requires realistic AI depictions in ads to be disclosed as "Generated by AI". Platform rules sit alongside, not instead of, the EU AI Act.

What is the difference between an AI Act provider and a deployer?

A provider builds or sells the AI system — model vendors, AI tool makers — and owes machine-readable marking plus detection means. A deployer uses AI and publishes AI-generated content — marketers, publishers, agencies — and owes labeling of deepfakes and AI text plus disclosure of AI interaction. One company can be both.

For guidance only — not legal advice. Based on Regulation (EU) 2024/1689 (EU AI Act), Article 50. Platform policies change — always check the current rules. v1.0 · October 2026 · reviewed quarterly.

⬇ Download the one-page PDF