Cabo Advisory

← Insights··Updated

The EU AI Act after the Omnibus — a short readiness check

Update (30 October 2026): The AI Omnibus (Regulation (EU) 2026/1744), in force since 27 July 2026, changed the timetable described below — the Annex III high-risk obligations now apply from 2 December 2027, and the Annex I product-embedded high-risk obligations from 2 August 2028. Most Article 50 transparency obligations kept their 2 August 2026 date. See our analysis of the Digital Omnibus. The readiness checklist below stands — with more time to execute it, and no excuse not to.

The AI Act entered into force on 1 August 2024, but its obligations arrive in stages. The prohibitions on unacceptable-risk practices have applied since February 2025, and the obligations for general-purpose AI models since August 2025. When this piece was written, the date that mattered for most companies was 2 August 2026, when the bulk of the regulation — including the high-risk regime of Annex III — was due to apply; the AI Omnibus has since moved the high-risk deadlines, as set out in the update note above.

A short readiness check:

  1. Inventory. Do you know, in one place, every AI system your company develops, procures or embeds? Most compliance gaps we see start with an incomplete inventory, not a misread article.
  2. Classification. For each system, determine the risk category — and document the reasoning. "Not high-risk" is a conclusion that needs a written basis, particularly where Annex III use cases (employment, credit, essential services, and others) are nearby.
  3. Role. The Act's obligations differ sharply between providers, deployers, importers and distributors. Many technology companies are more than one of these at once — for different systems, or even for the same system in different arrangements.
  4. Contracts. If you procure AI capability from vendors, your compliance depends partly on their conduct. Information obligations, cooperation duties and audit rights belong in the contract — not in a goodwill assumption.
  5. Governance. Transparency duties, human oversight, logging and post-market monitoring are operational processes, not documents. If they exist only in a policy PDF, they do not exist.

None of this needs to be dramatic. For companies that already run structured GDPR and information-security programmes, the AI Act is an extension of a familiar discipline — the work is in the inventory, the classification and the contracts. If any of those three are open questions for you, we should talk.