TL;DR
Article 4 of Regulation (EU) 2024/1689 applied to providers and deployers of AI systems on 2 February 2025. Member-State supervisory authorities started their work on 2 August 2026. From that date, national competent authorities can assess whether your staff have the literacy the regulation requires. The penalty regime in Article 99 became applicable on the same date. The obligation itself does not have a separate deferred deadline: it has been live since February 2025 and enforcement capacity is what changed in August 2026.
What the regulation actually says about timing
Article 4 of the AI Act sits in Chapter I of the regulation, the “general provisions” chapter. Per Article 113 of Regulation (EU) 2024/1689, the AI Act entered into force on 1 August 2024 and Chapter I (which contains Article 4) became applicable on 2 February 2025. The European Commission’s AI Act service desk confirms this date and lists the staggered application dates for the rest of the regulation.
The supervisory architecture is what arrived later. Member States had until 2 August 2026 to designate national competent authorities under Article 70. This is the date the enforcement clock started ticking in operational terms: from then on, a competent authority has the staff and the legal basis to ask you for evidence. The European Commission’s Digital Strategy AI Act page summarises the application dates and links the penalty articles.
Three dates to plan against
2 February 2025 - obligation live. Article 4 already applies to your organisation if you provide or deploy any AI system in the EU. You should already be able to point at a literacy programme that covers the staff who design, configure, monitor, and depend on AI outputs. If you cannot, you are out of compliance now, not at some future date.
2 August 2026 - supervision active. National regulators are now staffed and the AI Office at the European Commission is coordinating enforcement. From this date, requests for evidence from national authorities are realistic, especially for sectors already under regulatory scrutiny (finance, health, public administration). Portugal’s Government Digital Strategy names AI literacy as a prerequisite for the country’s AI deployment plans.
2 August 2027 - high-risk regime fully in force. Most high-risk AI systems become subject to the full AI Act regime, including the Annex III categories. If your organisation deploys or builds high-risk systems, the literacy you are documenting today underpins the conformity work you will need to evidence next year. Instituto Superior Técnico’s research on AI governance and curricula at Nova IMS both anchor on this date as the inflection point for high-risk deployment.
What “deadline” really means for you
Article 4 does not produce a single dated event you can put in a calendar invite and forget. It produces a continuous obligation: maintain a sufficient level of AI literacy in your workforce, evidence it on request. The pragmatic deadline is the next time a regulator, a customer in a regulated sector, or your own auditor asks for that evidence. That window opens fully on 2 August 2026 and stays open.
Plan for two checkpoints:
- Within 90 days of any new AI tool entering the workforce, run literacy sessions for the affected roles and log attendance.
- Annually, refresh role-tiered training and update the documentation pack you would hand a regulator on request.
How AISO Learn helps
Teams that need to land Article 4 training before a supervisory authority asks for it use our team-engagement programme - role-tiered curriculum, audit-defensible documentation, finished in eight weeks.
Talk to AISO Learn about Article 4 training