In this article8 sections
The date that mattered: 2 August 2026
EU AI Act enforcement August 2026 was a real milestone, but a narrower one than many articles predicted. On 2 August 2026, the Regulation, in force since 1 August 2024, began to apply in general, and national market surveillance authorities took over supervision. What did not happen on that date is the start of the high-risk rules. The Digital Omnibus on AI (Regulation (EU) 2026/1744, in force since 27 July 2026) moved those to 2 December 2027 for Annex III systems and 2 August 2028 for AI in products covered by Annex I.
Here is the full timeline as it stands now:
-
2 February 2025: prohibited practices (Article 5) and AI literacy (Article 4) apply.
-
2 August 2025: obligations for general-purpose AI model providers apply, and so does the penalties chapter. Member States were due to have their penalty rules in place.
-
2 August 2026: general application, including the Article 50 transparency rules. National market surveillance authorities supervise, including Article 4. The Commission can fine GPAI model providers.
-
2 December 2026: two new prohibitions (non-consensual sexual deepfakes and AI-generated child sexual abuse material), and the end of the grace period for machine-readable marking (Article 50(2)) for generative systems already on the market before 2 August 2026.
-
2 December 2027: high-risk rules for Annex III systems, such as recruitment, HR and credit scoring.
-
2 August 2028: high-risk rules for AI in regulated products (Annex I).
What actually changed on that date
The penalty rules themselves were not new on 2 August 2026: the penalties chapter has applied since 2 August 2025. What changed is that national authorities now supervise most of the Act. Article 99 of the AI Act sets the maximum fines:
-
Prohibited AI practices: up to €35 million or 7% of global annual turnover, whichever is higher.
-
A list of operator obligations, including those of providers and deployers of high-risk systems and the Article 50 transparency duties: up to €15 million or 3% of global annual turnover.
-
Supplying incorrect, incomplete or misleading information to authorities: up to €7.5 million or 1% of global annual turnover.
For SMEs and start-ups, each maximum is the lower of the two figures (Article 99(6)). Small mid-caps get the lower figure for the €15 million and €7.5 million tiers (Article 99(6a)). For an SME with €20 million in revenue, the €15 million tier caps out at 3%, or €600,000. Fines are set case by case, taking into account factors such as the nature, gravity and duration of the infringement and the size of the company (Article 99(7)).
These rules apply to providers and deployers. If you bought an AI tool from a vendor and use it at work, you are a deployer under the Act.
Who enforces this in the Netherlands
Each EU member state designates its market surveillance authorities. The Netherlands has not finished that step. The Dutch implementing law (Uitvoeringswet AI-verordening) is still a draft. The Autoriteit Persoonsgegevens (AP) is the intended coordinating supervisor, together with the RDI, but it says itself that it is not yet designated and only gets AI Act powers once the law is in force. Under the draft, sector supervisors such as the AFM and DNB take their own sectors, and Article 4 would be supervised by the AP, AFM and DNB.
In practice: in the Netherlands, no Dutch authority can fine you under the AI Act yet. That will change once the law passes, so the time to prepare is now, not when the first letter arrives.
Article 4: the obligation that already applies
Article 4 of the AI Act has applied since 2 February 2025. Article 4(1), as amended by Regulation (EU) 2026/1744, reads:
"Providers and deployers of AI systems shall take measures to support the development of AI literacy of their staff and other persons dealing with the operation and use of AI systems on their behalf, taking into account their technical knowledge, experience, education and training and the context the AI systems are to be used in, and considering the persons or groups of persons on whom the AI systems are to be used. This obligation does not require providers or deployers to guarantee any specific level of AI literacy of any individual."
This is not about training data scientists. It means the HR manager using an AI shortlisting tool, the compliance officer approving automated decisions, and the line manager receiving AI-generated performance scores should get training and guidance on what the system does, how it can fail, and when to override it. Article 4 does not set training hours, a test or a certification standard, and it does not require records. But records are how you show what you did, and the Commission suggests keeping an internal record of trainings.
The AI Act sets no specific fine for Article 4. Penalties are left to national law, and the draft Dutch law foresees an order to fix shortcomings (a "last onder dwangsom"), not a fine. For most SMEs, Article 4 is still the obvious first step, because it already applies to every company that uses AI.
The high-risk category you may meet, from December 2027
Annex III of the AI Act lists the AI use cases classified as high-risk. Point 4 is the one most relevant to companies with 20 to 200 employees:
AI systems intended to be used for the recruitment or selection of natural persons, in particular to place targeted job advertisements, to analyse and filter job applications, and to evaluate candidates.
If you use any tool that scores CVs, ranks candidates, or analyses video interviews, that is a high-risk AI system. A tool that infers candidates' emotions is not high-risk but prohibited outright (Article 5(1)(f)), as the Commission reads "workplace" to include recruitment. From 2 December 2027, the provider must carry out the conformity assessment and supply technical documentation and instructions for use (Article 13). As the deployer, you must then follow those instructions, assign human oversight to competent and trained staff, keep the logs the system generates for at least six months, and inform workers and candidates (Article 26).
Many HR teams are unaware that an off-the-shelf ATS with an AI ranking feature qualifies. The vendor is the provider, but you, as the deployer, have obligations of your own.
A preparation checklist
You do not need to rush, but you do need to start. Work through these steps in order.
Step 1: Map your AI systems. List every tool your company uses that contains AI, and mark the ones that automate or significantly influence a decision about a person. Include ATS platforms, performance management software, scheduling tools, and any chatbot used in customer service. Classify each one against Annex III and Article 5.
Step 2: Identify your role. For each system, determine whether you are a provider (you built it or put it on the market under your name) or a deployer (you bought and use it). Your obligations differ materially. Most SMEs are deployers.
Step 3: Train and document. Give every staff member who uses AI training that fits their tools and role, with extra depth for anyone in an HR or other high-risk use case. Keep a record of who did what and when. This is your Article 4 evidence.
Step 4: Check transparency. If you run a customer chatbot you built yourself, make sure it tells people they are talking to AI (Article 50(1)). If you publish deepfakes, label them (Article 50(4)).
Step 5: Ask vendors for documentation. For high-risk tools, ask now whether the vendor will deliver instructions for use, logging and conformity documentation by 2 December 2027. If a vendor cannot answer, that is a risk in your supply chain.
Step 6: Plan human oversight. For high-risk tools, decide who reviews AI outputs, with what authority, and how that is recorded. Article 26 requires this from 2 December 2027.
Step 7: Keep a gap log. Document what you have done, what gaps remain, and your plan to close them.
The one thing most companies skip
Most SMEs focus on the AI tools themselves and forget about internal governance. Appointing a named owner for AI, even if that person is also the COO or HR Director, creates accountability. Without a named owner, documentation stalls, training gets deprioritised, and vendor follow-up falls through the gaps. Forty-five minutes spent writing a one-page AI usage policy this month will make every subsequent step faster.
Document your AI literacy measures this month, and use the time until December 2027 to get your high-risk tools in order.
Not sure where your company stands? Run the free 2-minute compliance check at comply.khairos.ai to get a prioritised list of actions based on the AI tools you actually use.
References
More in this guide
Not sure where your company stands?
Answer 15 questions about how your team uses AI. You get a compliance score and the gaps to close first, as a PDF.