EU AI Act Article 50 is now in force: what chatbots, generative AI, and deepfake features must do

On 2 August 2026, Article 50 of the EU AI Act (Regulation (EU) 2024/1689) started applying. If your product includes a chatbot, a generative AI feature, or anything that produces synthetic audio, image, video, or text, this is no longer a "future compliance" item on a roadmap. It's active law, and the European Commission has published guidelines to clarify exactly what it covers.

It's worth reading past the headline, because the obligations split cleanly between two roles, and conflating them is the easiest way to under-comply.

If you're a provider (you build or supply the AI system), Article 50(1) and (2) apply to you directly. You must design any AI system intended to interact directly with a person so that the person is informed they're talking to an AI, unless that's already obvious from context. Separately, if your system generates synthetic audio, image, video, or text, you must mark the output in a machine-readable format that's detectable as AI-generated or manipulated. The text is specific that this marking has to be "effective, interoperable, robust and reliable as far as this is technically feasible," which leaves room for engineering judgment but not for skipping the requirement.

If you're a deployer (you operate someone else's AI system in your product or business), Article 50(3) and (4) apply. You must tell people when they're exposed to emotion recognition or biometric categorization. You must disclose when content is a deepfake. And if you publish AI-generated or AI-manipulated text on a matter of public interest without human editorial review, you have to disclose that too, unless a named person holds editorial responsibility for the content.

There are real carve-outs. Standard editing assistance that doesn't substantially alter input data or its meaning is exempt from the marking obligation. Artistic, satirical, or fictional works only need a disclosure that doesn't interfere with enjoying the work. Content reviewed by an editor with actual editorial responsibility is exempt from the public-interest text disclosure. The guidelines exist because these boundaries aren't self-evident, and the Commission clearly anticipated teams getting them wrong.

One detail that's easy to miss: this applies at "the latest at the time of the first interaction or exposure," per Article 50(5). That's a UX requirement, not just a legal one. Bolting a disclosure onto a settings page buried three taps deep does not satisfy this.

The stakes are not abstract. Non-compliance with Article 50 falls under the AI Act's penalty regime, which allows fines up to EUR 15 million or 3% of a company's total worldwide annual turnover, whichever is higher. That's the same order of magnitude as GDPR's most severe tier, and it applies regardless of company size.

For teams building software, mobile apps, or AI products with any EU-facing user base, this is now a concrete product requirement. If your app has a chatbot, an AI writing assistant, an image or voice generation feature, or anything touching emotion detection or biometric categorization, the practical work is: confirm disclosure appears at first interaction, confirm generated media carries a machine-readable watermark where the obligation applies, and confirm you can demonstrate compliance if a market surveillance authority asks. The Commission's guidelines point to the Code of Practice on Transparency of AI-generated Content as one accepted way to demonstrate compliance, worth evaluating before building a bespoke solution.

The practical takeaway isn't "add a small disclaimer and move on." It's that AI transparency now needs to be treated as a first-class product requirement, reviewed the same way accessibility or data-privacy requirements are, because as of this month, it carries the same kind of enforcement weight.

Source: Article 50, Regulation (EU) 2024/1689 (EU AI Act), and the European Commission's Guidelines on transparency obligations for providers and deployers of AI systems.