AI regulation in Europe is no longer a distant topic.
From 2 August 2026, the AI Act enters a new phase of enforcement and supervision, with powers held by the Commission and national authorities. Transparency obligations for certain AI systems have also started to apply.
This does not mean that every SME using ChatGPT, a chatbot or an automation suddenly has the same obligations as a company that builds AI models.
The regulation takes a risk-based approach and distinguishes between roles, uses and responsibilities.
But one conclusion matters for any business adopting AI:
it is no longer enough to ask whether the technology works. You also need to know how it is used, what it does in front of the customer, and what controls are in place.
What changed on 2 August 2026
From that date, the transparency obligations of Article 50 apply, and supervision of compliance rests mainly with national authorities.
At the same time, the AI Act's transparency rules start to apply.
Among them, certain interactive AI systems must tell people when they are talking to an AI rather than a human.
Some obligations also cover content that is artificially generated or manipulated in certain cases.
The underlying idea is simple: people must be able to tell when they are dealing with an AI system in situations where that matters.

1. If a customer talks to an AI, transparency matters
This matters a lot for My NovaGen AI, because we build agents and chatbots.
If someone starts a chat thinking they are talking to an employee, when an AI is handling everything, the experience can be misleading.
Chatbots and other affected interactive systems must tell users they are dealing with an AI.
This also fits good product practice:
there is no need to hide automation for it to work well.
A simple line such as:
“I’m Nova, the virtual assistant of My NovaGen AI”
sets the right expectations without breaking the conversation.
2. Transparency does not mean filling the chat with legal notices
Meeting transparency duties should not turn a simple experience into a wall of warnings.
Show the information where it makes sense.
For example:
- at the start of an automated conversation;
- before any interaction where the user could reasonably think they are talking to a person;
- when AI-generated or AI-altered content is shown in cases covered by the labelling rules.
Good transparency is part of the user experience.
It is not fine print added at the end.
3. An SME needs to know what role it plays
The AI Act distinguishes between different actors.
A company may:
- develop an AI system;
- provide a system;
- deploy or use a third-party system;
- integrate a general-purpose model into its own product.
The obligations differ from case to case.
That is why an SME using an existing service should not just copy a checklist written for a model provider.
The first question is:
what are we actually doing with AI?
4. Not all obligations have the same date
The AI Act applies in stages.
Application is phased, with different dates for each obligation.
Some obligations already applied before August 2026, such as those on certain prohibited practices and AI literacy.
Other rules, especially some on high-risk systems, come later. According to sources consulted, after the 2026 review (digital omnibus), the obligations for high-risk systems under Annex III will apply from 2 December 2027, and those under Annex I from 2 August 2028. Confirm this against the official text.
So avoid summing up the regulation as “everything is mandatory from August 2026”.
That is not accurate.
Check the type of system and the specific provision.

5. What an SME should do right now
This is not specific legal advice, but a few practical steps make sense.
Know where we use AI
Not just tools. Processes.
For example:
- website chatbot;
- WhatsApp agent;
- content creation;
- lead classification;
- internal analysis;
- draft generation.
Identify what the customer sees
Does the AI answer directly?
Does it generate public content?
Does it take actions for the business?
Define responsibility
Who reviews how it behaves?
Who can change its instructions?
Who handles an incident?
Document limits
What it can and cannot do.
Review transparency
The person must know when they are dealing with an AI system, if the case falls under the applicable obligations.
6. Regulation reinforces an idea that was already good engineering
Many of these principles make sense even before any legal question comes up.
A good business system can explain:
- what it does;
- what data it uses;
- who is responsible;
- when a person intervenes;
- how an error is corrected;
- what gets recorded.
Regulation is pushing AI from a “black box” to a managed system.
That is good news for serious businesses.
7. The mistake would be to react with fear or with indifference
There are two unhelpful extremes.
“We can no longer use AI”
That is not the point of the regulation.
The EU explicitly aims to balance innovation with protection against risks.
“This only affects big tech”
Not at all.
An SME may have obligations depending on how it uses certain systems.
The sensible response is to understand the use case and apply proportionate measures.

8. What it means for agents and chatbots
If your business uses a customer service agent, a mature setup should cover at least:
- Clear identification of the assistant.
- Approved knowledge.
- Limits on what it can do.
- Handover to a human.
- Activity logging.
- Data protection.
- Supervision of its behaviour.
- Review when the model, instructions or processes change.
These measures do not replace a legal review when one is needed.
But they help make compliance part of the design, not a last-minute fix.
9. Important news for the sector, and for customers
This new enforcement phase changes the commercial context.
Customers will increasingly ask:
- does this use AI?
- what information does it consult?
- what happens to my data?
- can I speak to a person?
- who controls the answers?
Answering these questions well can be a real competitive advantage.
Not because the company has “more compliance”, but because it shows it understands what it has built.
Practical summary for an SME
In practice: before adding more AI, know which systems you use, what the customer sees, what data is involved, and who keeps control when something unexpected happens.
| Question | What to review |
|---|---|
| Does the person know they are talking to AI? | Clear identification of the assistant, where applicable. |
| What role does the company play? | Whether it develops, integrates or provides the system, or simply uses a third-party one. |
| What data does the system use? | Necessity, minimisation, access and accountability for that data. |
| What happens if the AI makes a mistake? | Human handover, logging, correction and review of behaviour. |
Frequently asked questions
Do all SMEs using AI have the same obligations?
No. The AI Act distinguishes between types of systems, risks and roles. Using a third-party tool is not the same as developing or providing a model. The answer depends on the specific use case.
Must a chatbot say that it is an AI?
The transparency obligations of Article 50 apply in certain cases. Beyond any legal duty, clearly identifying the assistant is usually good user experience practice.
Does this replace a legal review?
No. This article offers a practical reading for businesses, but a specific case may need specialist legal advice.
If you are reviewing how AI is presented to your customers, take a look at our WhatsApp AI agents and AI web chatbots.
Conclusion
August 2026 is an important milestone for AI in Europe.
The regulation moves into a more operational phase, and transparency obligations are no longer a future issue for certain systems.
For an SME, the message should not be “we need more paperwork”.
It should be:
we need to know which AI we use, what it does, what the customer sees and who keeps control.
That discipline helps with both compliance and building better systems.
This article is for information only and does not replace specific legal advice on a particular case.
Reference sources
Information gathered from specialist press and legal analysis published after 2 August 2026. We have not checked it against the official text of Regulation (EU) 2024/1689. This article is general information, not legal advice. The application date of each obligation depends on the type of system.