EU AI Act Article 50: What Agencies Actually Need to Know

Article 50 of the EU AI Act is the part of the law aimed at transparency: making sure people know when they're dealing with AI, and when content has been artificially generated. It entered into force on August 2, 2026, with a grace period for some marking requirements running to December 2, 2026.

Most of what's been written about it focuses on big platforms — the ChatGPTs and Midjourneys of the world. That's fair: the heaviest obligations (marking AI outputs in machine-readable format) fall on the companies that build AI systems, not the agencies that use them. But Article 50 also creates obligations for deployers — anyone using an AI system in a professional capacity — and that includes agencies and freelancers.

The four things Article 50 actually covers

For most agency work — ad copy, product descriptions, social captions, design mockups — the first three rows don't apply to you directly; you're not building the AI system or running biometric software. The fourth row is the one worth understanding closely.

The exemption that matters most: human editorial review

Article 50(4) requires deployers to disclose AI-generated or manipulated content when it concerns matters of public interest — think news, politics, public health, public safety. But it carves out an exemption:

Two things follow from this. First, most day-to-day agency output (brand marketing, product content, internal comms) likely doesn't fall under "public interest" content to begin with. Second, even where it might — a client in journalism, public health, or civic advocacy, for example — a real human review step can exempt you from the labeling requirement entirely.

Who's actually on the hook, and for how much

Enforcement sits with EU authorities, and fines can run up to €15 million or 3% of global annual turnover, whichever is higher. In practice, enforcement attention so far has concentrated on large AI providers and platforms rather than small agencies. That said, the deployer obligations in Article 50 are written broadly enough to apply to any business using AI professionally if its output reaches EU users — so "we're too small to matter" isn't a compliance strategy, just a bet on enforcement priorities.

Beyond the EU: why this matters regardless

Even where Article 50 itself doesn't strictly require disclosure, there are other reasons agencies are moving this direction anyway:

• Client contracts increasingly ask for it. More RFPs and SOWs now include AI-usage disclosure clauses, independent of what any specific law requires.
• US state laws are moving too, unevenly. Texas's TRAIGA (in force since January 1, 2026) requires disclosure of AI interaction in some contexts. Colorado's original AI Act was repealed and replaced by a narrower law that takes effect January 1, 2027. California's SB 942 mostly targets large AI platform providers (1M+ monthly users), not agencies using third-party tools — but that could change as amendments move through. The state landscape is genuinely inconsistent right now, which is exactly why having your own documentation habit, rather than chasing each jurisdiction, is the practical move.
• The FTC has been active on undisclosed AI use in advertising and endorsements under existing deceptive-practices authority, separate from any AI-specific law.

The common thread: whether or not a specific statute forces your hand, being able to produce a clear record — what AI tool was used, on what, and who reviewed and approved it — is the thing that actually protects you, in a client dispute, a regulator inquiry, or just a client who asks.