EU AI Act Transparency Rules Are Now Live and Enforceable

15 million euros, or 3 percent of global annual turnover, whichever is higher. That is the ceiling now facing companies that ignore the European Union AI Act's transparency obligations, which took effect on August 2, 2026.

The provision that switched on is Article 50, and it lands in three places:

• Chatbots and conversational systems: users must be told they are dealing with an AI, unless that is already obvious from context.
• Synthetic content: AI-generated images, audio, video and text must carry machine-readable markings.
• Emotion recognition and biometric categorisation: the people subject to these systems must be informed.

The part that slipped

The heaviest section of the Act did not arrive on schedule. Full compliance for the high-risk systems listed in Annex III, covering biometrics, employment and hiring, education, access to essential services, and migration and border management, has moved to December 2, 2027.

The reason is procedural rather than political: the technical standards and guidance that compliance would be measured against were not ready. Enforcing a rule nobody can yet demonstrate conformity with helps no one.

Seen end to end, the Act's staged design is clear. Prohibited practices and AI literacy came into force in February 2025, general-purpose AI model rules in August 2025, transparency in August 2026, and high-risk systems in December 2027.

Does this reach companies outside the EU?

For many of them, yes. The Act keys on where a system is used, not where its vendor is incorporated. A support bot on a store that ships to Germany, an AI assistant inside a SaaS product with European customers, AI-generated visuals in a campaign running in the EU: all of it sits inside the frame regardless of the company's home market.

The most common gap we run into is the chat bubble in the corner of a website. Those bots are usually given a human name and a friendly avatar, and nowhere in the interface does it say a machine is answering. Under Article 50 that is now a defect to fix, not a design preference.

The second gap is marking. "Machine-readable" does not mean a caption under the image saying it was AI-generated. It means a signature embedded in the file itself that software can detect, and for most content teams that step does not exist in the production pipeline at all today.

A practical starting point

Before wading into the legal text, map your own exposure with three questions. Do you have users in the EU? Do you run an AI interface that talks to them? How much of what you publish is machine-generated? Three noes means nothing is urgent for you this month.

For every yes, the remediation is smaller than the penalty figure suggests: a clear disclosure line in the bot interface, a signing step in the content pipeline, and enough record-keeping to show your work. Our advice is to treat it as routine product maintenance rather than a compliance project. The high-risk obligations landing in December 2027 are far more demanding, and teams that close the transparency gap properly now will be in much better shape when that wave arrives.

Sources: European Commission, Al Jazeera, Cooley