Transparency obligations are separate from high-risk

You might have less time than you think to meet transparency requirements - as do your agents

Consultation deadline: 3 June 2026

The European Commission released draft guidelines on Article 50. As with the recent high-risk guidelines, the Commission is doing a great job of clearing up myths.

A common myth I hear about the AI Act is “there are four risk classes”. Probably because this is the answer you get from Copilot:

But this division is misleading. The Act never uses the categories “limited risk” or “minimal risk”, and in any case, these categories aren’t mutually exclusive. 

The draft transparency guidelines just clarified:

AI systems falling within the scope of Article 50 AI Act [with transparency risks] may also be classified as high-risk and be subject to the relevant requirements and obligations for such systems.

(point 22, Guidelines on the implementation of transparency obligations)

In other words, your high-risk system can also have transparency risks - and only high-risk obligations have been delayed.

• 2 August 2026: Immediate transparency requirements if you ship your system on or after this date.

• 2 December 2026: Transparency requirements if your system is already on the market.

Almost everyone is working with an AI system that carries a transparency risk: 

1. interacting with the AI system - people have to know this 

2. the system is generating synthetic context (e.g. text) - this context has to be marked

3. recognizing emotions - remember that this is prohibited if at a workplace or in a school 

4. generating text to inform the public on matters of public interest - relevant to quite a lot of public sector AI use 

5. deepfakes (combined with #4 in the Act, while I’ll never understand)

If you are already providing an AI system that meets any one of these criteria, you have until 2 December to meet transparency requirements.

AI agents aren't exempt

AI agents also have transparency requirements, if people will interact with them. As the Commission has said previously, a provider cannot throw their hands up and claim product flexibility as a reason to not anticipate behavior. If the agent may interact with a person, then the provider has transparency requirements.

Where the provider cannot reliably determine whether the AI agent will interact with a natural person, the agent should be instructed to disclose itself as such in every situation where it is likely that the agent may interact with a natural person.

(point 28)

The guidelines give specific examples of insufficient transparency:

- Disclosures contained only in terms and conditions, URLs or documentation - Generic references to "assistant" - Statements solely referring to underlying technologies (e.g. “this system uses LLMs”) without explaining the function or

(point 35)

If you a provider and want to check your deadlines, to confirm you're meeting your requirements - check out our services for developers.