In this article10 sections
- AI Act Enforcement in 2026: The Signal Behind the Noise
- The Regulatory Calendar That Actually Matters
- What Changed on 2 August 2026, and What Didn't
- What National Authorities Have Actually Said
- The GDPR Precedent: Guidance First
- Which Sectors to Watch
- Sizing Your Compliance Investment Correctly
- Penalties Are Real, But Context Matters
- One Action to Take Before the End of This Quarter
- References
AI Act Enforcement in 2026: The Signal Behind the Noise
For a long time, 2 August 2026 was sold as the day the AI Act would "bite". Then the AI Omnibus (Regulation (EU) 2026/1744, in force 27 July 2026) moved the high-risk rules back. So what actually changed on 2 August 2026, and what didn't?
The short version: national supervision started, the transparency rules began to apply, and the high-risk rules are still more than a year away. That is good news for SMEs. It means you can spend your compliance budget in the right order.
The Regulatory Calendar That Actually Matters
The AI Act entered into force on 1 August 2024. From there, obligations phase in:
-
2 February 2025: The prohibitions in Article 5 apply. They include social scoring (by anyone) and real-time remote biometric identification in public spaces for law enforcement, with narrow exceptions. The AI literacy obligation in Article 4 has applied from the same date.
-
2 August 2025: Rules for general-purpose AI (GPAI) model providers apply.
-
27 July 2026: The AI Omnibus enters into force. It rewrites Article 4 and moves the high-risk dates.
-
2 August 2026: General application of the Act. Article 50 transparency duties apply, national market surveillance authorities supervise Article 4, and the Commission can fine GPAI model providers.
-
2 December 2026: Two new prohibitions apply: AI that creates non-consensual sexual deepfakes, and AI-generated child sexual abuse material.
-
2 December 2027: High-risk rules for Annex III systems (such as recruitment and credit scoring), including deployer duties (Article 26) and the fundamental rights impact assessment (Article 27).
-
2 August 2028: High-risk rules for AI in products covered by Annex I.
What Changed on 2 August 2026, and What Didn't
What changed. Supervision moved from paper to practice in the Member States that have designated their authorities. Germany's KI-Marktüberwachungsgesetz entered into force on 29 July 2026, with the Bundesnetzagentur as the default authority. Article 50 now applies: chatbots must disclose they are AI, and deep fakes must be labelled.
What didn't change. Deployers of high-risk systems have no Article 26 duties yet; those start on 2 December 2027. And in the Netherlands, the implementing law is still a draft. The Autoriteit Persoonsgegevens (AP) is the intended AI supervisor, but it states itself that it has no AI Act powers until that law is in force.
What National Authorities Have Actually Said
The clearest statements come from the Commission and the AP:
-
The Commission's AI literacy Q&A says national market surveillance authorities supervise Article 4 from August 2026, that any sanction must be proportionate, and that a sanction "might, however, be more likely if there is proof of an incident due to lack of appropriate training and guidance".
-
The AP (9 October 2026) confirms it is the intended market surveillance authority for AI but is not yet designated.
-
The explanatory memorandum to the draft Dutch implementing law says a shortfall in AI literacy leads to an order to fix it (a last onder dwangsom), not a fine.
We are not aware of any fine for Article 4 so far.
The GDPR Precedent: Guidance First
GDPR went fully applicable on 25 May 2018. In the period that followed, most supervisory resources went into investigations, consultations, and guidance. Fines were real but concentrated, and the early ones targeted organisations with no visible compliance effort.
AI Act supervision is likely to follow a similar arc. National authorities are still being designated and resourced. The European Commission's digital strategy framework describes a system in which the AI Office supervises GPAI model providers and national authorities supervise everyone else.
What this means in practice: organisations that can show a documented, good-faith effort are in a strong position. Organisations with no records, no transparency notices, and no staff training are the ones that will struggle to answer a supervisor's questions.
Which Sectors to Watch
Based on the structure of Annex III, three sectors stand out once the high-risk rules apply:
HR and recruitment. Automated CV screening and performance monitoring tools are listed in Annex III. If your company uses an AI-assisted hiring tool, from 2 December 2027 you have deployer obligations around human oversight, logging and informing workers.
Financial services. Credit scoring of individuals and life and health insurance pricing appear in Annex III and are already subject to existing financial supervision.
Healthcare. Emergency patient triage is listed in Annex III. Diagnostic AI is usually a medical device; where the device needs a notified-body assessment, it is high-risk through Annex I from 2 August 2028.
Sizing Your Compliance Investment Correctly
Many companies in 2018 either dramatically over-invested in GDPR or did almost nothing. The sweet spot was proportionate, documented effort. For an SME under the AI Act, that means four concrete things:
-
An AI system inventory. Know what AI tools you are using, which could qualify as high-risk under Annex III, and which vendor supplied them. You cannot plan for Article 26 without this baseline.
-
AI literacy measures. Article 4 requires you to take measures to support your staff's AI literacy; no specific level is required. For SMEs, role-appropriate training is the usual measure. Records are how you show the measures you took.
-
A transparency mechanism. Article 50 requires that people interacting with a chatbot are told it is an AI, and that deep fakes are disclosed. If you use these tools, check that the notice is in place.
-
A fundamental rights impact assessment, only if it applies to you. If you are a public body, provide a public service, or score credit or price life and health insurance, Article 27 will require a FRIA from 2 December 2027. Most private SMEs never need one.
A structured programme over two to three months, with proper documentation, is realistic for a 50-person company.
Penalties Are Real, But Context Matters
The AI Act sets maximum fines of €35 million or 7% of global annual turnover for prohibited practices under Article 5, and €15 million or 3% for most other violations, including deployer obligations under Article 26 (from 2 December 2027). For SMEs and start-ups the lower of the two amounts applies (Art 99(6)). The Act sets no specific fine for Article 4; penalties are left to national law. Actual fines must be proportionate and take into account, among other things, the severity of the violation and the measures the organisation took (Article 99).
For an SME with €5 million turnover, a 3% cap is €150,000. Serious, but far from the headline figures.
One Action to Take Before the End of This Quarter
If your organisation has not yet mapped its AI tools against the Annex III high-risk categories, start there. It takes less time than most compliance officers expect, and it is the foundation for everything else: your AI literacy measures, your vendor conversations, and your plan for December 2027.
Run the free 2-minute compliance check at comply.khairos.ai to see which AI Act obligations apply to your organisation and where your gaps are likely to be.
References
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.