Live since February 2025
The Article 5 prohibited practices and the Article 4 AI literacy duty. Both bind providers and deployers today.
EU AI Act readiness
The final Digital Omnibus moved the standalone high-risk obligations to 2 December 2027, and embedded systems to August 2028. The prohibited-practice ban and the AI literacy duty have applied since February 2025, the penalty framework since August 2025. We map where your AI footprint stands against the Act and what to do first.
What applies, and when
The Article 5 prohibited practices and the Article 4 AI literacy duty. Both bind providers and deployers today.
The penalty framework under Articles 99 and 100, the governance bodies including the AI Office, and the obligations on general-purpose AI model providers.
The Article 50 transparency duties and the innovation-support measures. Content marking under Article 50(2) is deferred for systems already on the market.
The standalone Annex III high-risk obligations, moved back by the final Digital Omnibus.
High-risk AI that is a product, or a safety component of a product, covered by Annex I harmonisation legislation.
The Council adopted the final Digital Omnibus on 29 June 2026. A proposal that still sells you a high-risk compliance deadline this year is working from dates that no longer apply. The penalty framework it points at has been in force since August 2025.
The honest picture
Obligations attach to your role for each system, not to the company as a whole. A firm using an AI feature inside a SaaS tool is a deployer, carrying a much lighter set of duties than the provider who built it. The job of an assessment is to establish that most of the footprint sits in the minimal and limited tiers, then find the few systems that do not.
Employment and worker management covers recruitment, screening, task allocation, monitoring, evaluation and promotion. Access to essential services covers creditworthiness and credit scoring, life and health insurance pricing, and benefits eligibility. Those two categories account for most mid-market exposure.
Article 6(3) takes an Annex III system out of high-risk when it performs a narrow procedural or preparatory task. Any system that profiles natural persons stays high-risk regardless, the basis has to be documented before deployment, and an undocumented derogation claim is itself a gap.
Putting your own name on a high-risk system already on the market, or substantially modifying one, makes you its provider under Article 25. The heavier obligations move onto you. This is the most expensive surprise in a deployer engagement.
Exposure, stated accurately
Article 99 sets three tiers. The 35 million euro or 7% ceiling applies to the Article 5 prohibited practices. Ordinary operator and deployer breaches sit at 15 million or 3%, and supplying incorrect or misleading information to authorities at 7.5 million or 1%. For SMEs and start-ups, Article 99(6) makes the fine the lower of the cash amount and the percentage rather than the higher, and most mid-market companies qualify. Quoting the top tier to a board overstates what you actually face, and someone in that room usually knows it.
Three things worth doing now
One row per AI tool or feature in use across departments, with a named business owner. Everything else keys off it, and more than half of mid-market firms have none.
Sixty to ninety minutes for the staff who operate AI. This discharges the live Article 4 duty and is the cheapest control you can show anyone who asks.
A line telling people they are dealing with an AI assistant, and a labelling rule for AI-generated content. Near zero cost, and it removes the most visible gap you have.
The assessment
Every system across departments, shadow AI included, with the role and risk tier assigned per system and a written rationale behind each one.
Ten dimensions scored, from classification correctness and human oversight through to vendor documentation, logging and the GDPR interface. Anything scoring three or below is flagged for remediation.
A twelve-month plan sequenced by risk and effort, a memo your team can start on the same week, and a briefing the board can read.
We cap scope at ten to fifteen systems. Deliverables, timing and a fee are agreed after the first conversation.
We map systems against the Act and give your team a prioritised plan it can act on. Binding classification of high-risk or borderline systems, conformity assessment and CE marking are functions for an EU AI Act lawyer and a notified body. We work alongside yours, or help you find one.
Regulatory position reviewed 10 July 2026 against Regulation (EU) 2024/1689 and the final Digital Omnibus on AI. Dates are re-verified at the start of every engagement.
Let’s talk
A 20-minute conversation about your operating priorities and one real bottleneck. We can identify where AI could help and what a useful first step would involve.