Here's a date you may have missed amid the summer holidays: 2 August 2026. That's when Article 50 of the EU AI Act took effect. Forget the headline-grabbing "high-risk AI" rules everyone's been talking about for two years; those only catch a narrow slice of use cases. Article 50 is the provision most businesses are likely to encounter because it applies to several common uses of AI, not just high-risk systems. It applies the moment your AI does one of four things: talks directly to a person, generates synthetic audio, image, video or text, does emotion recognition or biometric categorisation, or produces a deepfake or AI-written text aimed at informing the public. If you've used AI for anything customer-facing or content-related lately, there's a decent chance it applies to you.

But we're British - surely this doesn't apply to us?

Article 2 of the AI Act was drafted precisely to stop that argument working. It reaches providers and deployers anywhere in the world, including the UK, the moment the output of their AI system is used in the EU. No Brexit escape hatch, no equivalence deal, no asterisk for British business. A firm with zero EU staff, zero EU offices and zero EU servers can still be caught, purely because the people on the other end of its chatbot, or reading its content, happen to be in the EU. And there's no UK law standing in to soften the blow: our own approach to AI regulation is still principles-based and largely voluntary, so it offers no shortcut around the AI Act where the AI Act genuinely applies.

So when do you actually have to put your hand up and say "a robot wrote this"?

This is the bit that is likely to matter most for day-to-day work: Article 50(4). In plain English, it says that if you're using AI to generate or tweak certain content, you may have to tell your audience it's AI-assisted. Whether you actually do depends entirely on what you've made and who's reading it. Let's run through a few real-world examples:

Letter, Email or Message drafted with AI help, then tidied up by you? That letter is private correspondence to one named recipient. Article 50(4)'s labelling duty will ordinarily not apply because private correspondence is not published to inform the public on matters of public interest. Same goes for internal memos and anything else that stays behind closed doors.

What about that punchy AI-written ad copy? Generally, still fine for Article 50(4) purposes - a product blurb or campaign slogan isn't "informing the public on matters of public interest" in the sense the law means, even if a chatbot wrote every word. But don't get too comfortable: if your marketing includes an AI-generated image, video or voiceover that could pass for the real thing - think a synthetic spokesperson or a manipulated product demo that's a different obligation entirely (the deepfake rule), and it applies regardless of how harmless the context feels.

Social media posts - This is where things get genuinely tricky, because it all comes down to what you've posted and why. A synthetic photo or clip that appears to depict a real person, place or event and would reasonably be taken as authentic may constitute a deepfake, wherever you posted it. AI-written commentary on something that actually matters to the public - politics, health, the economy, the environment - can also trigger the labelling duty, even on an informal platform, as long as it's reaching an open audience rather than a private group chat.

Articles, blogs etc - the professional article, the client briefing, the blog post you're reading right now. This is exactly the territory Article 50(4) was designed to address: published content intended to inform the public on matters of public interest. If that content is AI-generated or materially AI-manipulated, you may need to label it, unless a real human has substantively reviewed it and an identifiable person takes editorial responsibility for what is published. And "substantively reviewed" means just that: mere superficial review is unlikely to be sufficient, whereas a substantive rewrite, fact-check and sign-off by a named author is far more likely to satisfy the exemption. If you can't honestly tick that box, the fix is easy - just say so, clearly and near the top, not buried in the small print.

What's coming next, and when

Article 50 forms part of a wider implementation timetable under the AI Act. The prohibited practices under Article 5 took effect on 2 February 2025. Obligations for providers of general-purpose AI models followed on 2 August 2025. Article 50 then took effect on 2 August 2026, subject to one transitional provision: providers of a generative AI system already on the market before that date have until 2 December 2026 to implement the machine-readable marking of their outputs, a separate, provider-side obligation from those addressed above. The AI Act's remaining high-risk requirements, including risk management and technical documentation obligations, will apply from 2 December 2027 for standalone high-risk use cases, and from 2 August 2028 for AI embedded in products already regulated under other EU law.

What’s the Penalty for Getting it wrong?

Very much like the GDPR, the EU AI Act has real teeth. Serious breaches of the transparency obligations can attract fines of up to EUR 15 million or 3% of worldwide annual turnover, whichever is higher.

The bottom line

Article 50 isn't trying to stop you using AI. It just wants you to be upfront about it, and how upfront depends on what you've made and who's watching. If your business talks to, markets to, or publishes for anyone in the EU, "we're a UK company" isn't the shield it used to be. The smart move is a quick audit: work out what you're publishing that touches the public interest, where synthetic media might be creeping into your content, and where a named human genuinely stands behind what's gone out the door. Do that now, and there'll be no awkward surprises later.

To find out more please contact a member of our Intellectual Property Team.