Insights · EU AI Act

31 JULY 2026

Is your organisation ready for the new transparency obligations?

As of Sunday, new transparency obligations under the EU AI Act apply.
No reason for panic – but a good moment to establish what this means for your organisation in practice.

The EU AI Act is not a law that switches “on” in a single day. It comes into effect in phases, which makes it difficult to keep track of what applies and when. On 2 August 2026, the transparency obligations under Article 50 come into effect: rules on AI chatbots, AI-generated content and deepfakes, which apply regardless of whether an organisation considers itself an “AI company”.

As the deadline approaches, several misconceptions continue to circulate, much like those we saw during discussions surrounding NIS2 (Cyberbeveiligingswet). Below are the four we encounter most often in conversations with SME organisations – and what actually changes under the regulation itself.

Myth 1

“The AI Act has been delayed anyway”

This is partly true, which is precisely what makes the situation confusing. In June 2026, the European Parliament and the Council of the EU reached agreement on the so-called Digital Omnibus on AI. This gives organisations 16 months’ extra time for the heaviest category: high-risk AI systems such as recruitment software, credit scoring or biometric identification. The new deadline for these is 2 December 2027, instead of August 2026. For AI embedded in existing products, a comparable 12-month deferral applies.

That deferral only concerns high-risk systems. The transparency obligations under Article 50 – the subject of this article – were left untouched in that negotiation and simply take effect from 2 August 2026, regardless of whether an AI system is classified as high-risk.

Reality

Only the heaviest part of the law has been deferred. The transparency obligations of 2 August 2026 remain in force and can affect a large number of organisations that use AI.

Myth 2

“This only affects companies that develop AI themselves”

The term “AI Act” can easily give the impression that the legislation is aimed primarily at technology companies developing AI models.

The obligations apply to two roles: providers of an AI system, and organisations that deploy AI systems (the so-called “users” or “deployers”). An SME that deploys a ready-made chatbot on its own website, or has content generated using an external AI platform, is itself a deployer – with its own responsibility, regardless of who built the underlying model.

Reality

Whoever uses AI has their own obligations – even without ever having written a single line of code themselves.

Myth 3

“We don’t use AI systems”

Many business owners think of an “AI system” as something highly advanced – a self-learning platform or a proprietary model. In practice, it’s often about tools that have already been in use for months.

Article 50 covers four concrete situations that are common among SMEs. A chatbot or virtual assistant on a website must make clear that the visitor is dealing with AI, unless this is obvious from the context. Certain forms of AI-generated or AI-manipulated content are subject to transparency or marking obligations. Systems that recognise emotions or categorise people based on biometric characteristics require a notification to the person concerned. And AI-generated text on matters of public interest, or a deepfake, must be labelled as such.

One nuance: for generative AI systems already on the market before 2 August 2026, a transitional period until 2 December 2026 applies specifically to the machine-readable marking requirement. The other elements of Article 50 – recognisability towards people, the notification duty for emotion recognition – do simply apply from 2 August.

Reality

A chatbot on your website or AI-generated marketing content may well fall within the scope of Article 50. The question isn’t whether you use AI, but where.

Myth 4

“AI literacy is still a long way off”

Of all the obligations in the AI Act, Article 4 – AI literacy – is the one most often overlooked, probably because it’s the least conspicuous.

This obligation has already been in force since 2 February 2025. Since the Digital Omnibus, it has been made explicit that this is an obligation of effort, not a prescribed training programme or an obligation of result. Organisations must take demonstrable measures to ensure that staff and contractors, regardless of seniority, sufficiently understand how the AI they use works and what the associated risks are, with those measures calibrated to their role and the context in which AI is deployed.

This reflects a pattern we encounter regularly: AI rarely enters an organisation through a board decision. It enters because an employee finds a tool useful. Research into business AI use shows that some of these prompts contain more sensitive information than necessary – usually from ordinary, well-intentioned use, not malicious intent. That makes AI literacy less a compliance checkbox and more a question of visibility: knowing which tools are being used, and for what.

Reality

AI literacy has been mandatory since February 2025. The practical question isn’t whether, but whether you can demonstrate which measures you’ve taken.

Conclusion

Not a cause for panic, but a moment for action

In practice, the AI Act is often viewed as a distant issue that will somehow resolve itself over time. That does the law no justice. Two elements are already in effect: the transparency obligations that apply from 2 August 2026 and the AI literacy requirements that have been in force since February 2025. Both can be managed with common sense – it starts with knowing where AI is used in your organisation, and by whom.

Looking for a quick assessment of your current position? We’ve built a short, free check for exactly that.

Take the free EU AI Act Preparation Check →

Six questions, three minutes – nothing is submitted or stored.

Already have a clear picture and would like to discuss your organisation’s position?

Sources