On 2 August 2026, the transparency obligations under Article 50 of the AI Act become applicable. Much else has been postponed. Here is, soberly, what actually applies and what does not.
The AI Act enters into force in stages. On 2 August 2026, it is above all the transparency obligations under Article 50 that become directly applicable. They concern the use of certain AI systems, regardless of company size or industry:
Anyone operating a chatbot or voice assistant must ensure that users can recognise they are interacting with an AI system. This does not apply where it is obvious from the circumstances.
Artificially generated or manipulated image, audio and video content (such as deep fakes) must be labelled as such. Disclosure obligations also apply to certain AI-generated texts that inform the public on matters of public interest.
Where systems for emotion recognition or biometric categorisation are used, the persons concerned must be informed.
For providers of general-purpose AI models (GPAI), enforcement of the obligations already in force begins.
Not everything currently reported about the AI Act applies from August. Two large blocks have been moved back:
Stand-alone high-risk systems (Annex III), for example AI in recruitment, credit scoring or critical infrastructure.
AI in regulated products (Annex I), for instance in machinery, medical devices or vehicles.
Important: The Article 50 transparency obligations are largely unaffected by this postponement. One exception applies to the labelling of AI-generated content already in circulation before 2 August 2026. A transitional period runs until 2 December 2026 for that. Anyone operating a chatbot today or publishing new AI content could, however, already be under obligation from 2 August 2026.
The Article 50 obligations could already apply where a company:
On fines: Article 99(4) of the AI Act provides for fines of up to €15 million or 3% of worldwide annual turnover for breaches of the transparency obligations, whichever is higher. For SMEs including start-ups, the lower of the two amounts applies (Article 99(6)).
Both frameworks are part of the same development: the EU increasingly requires companies to show demonstrable digital diligence, in cybersecurity (NIS2) as in the use of AI (AI Act). Companies affected by NIS2 frequently use AI tools as well and should keep both sets of obligations in view.
To be honest: NIS2 Pilot is an app for your NIS2 preparation, not an AI Act tool. If you want to know where your company stands on NIS2 implementation, the free check will help you within minutes.
The transparency obligations contain no general exemption for small and medium-sized enterprises. What matters is the use of the respective AI systems, not company size. Size is, however, taken into account when fines are assessed.
Under the wording of Article 50, the information must reach the persons concerned in the context of the interaction or content, at the latest at the first interaction. A hidden notice is unlikely to suffice in many cases. How supervisory practice will interpret this in detail will become clear after the deadline.
The applicability of the high-risk obligations under Annex III was moved to 2 December 2027. That buys implementation time. But the requirements (risk management, data governance, documentation, human oversight) are extensive, so early preparation is likely to be worthwhile.
