Market Surveillance and Enforcement Under the EU AI Act (Articles 74–81)

How national market surveillance authorities police the AI Act day to day: who has jurisdiction over which systems, the corrective-action procedure that follows a risk finding, and the Union safeguard procedure that resolves disagreements between Member States.

Key context
"Market surveillance authorities have significant powers: they can request documentation, conduct testing, order withdrawals, and impose provisional measures — all without prior court authorisation in urgent cases."
EU AI Act, Article 74
A customs inspector's stamp poised over a shipping manifest, evoking the market surveillance authority's power to halt a non-compliant product at the point of market access.

Builds on AR-000142 (Human Oversight), AR-000210 (High-Risk AI Systems), and AR-000270 (GPAI Obligations) — none of their content is re-derived here. AR-000270's coverage of GPAI enforcement (Articles 88–94) is a structurally separate track from the Articles 74–81 mechanism covered here.

Verification-depth disclosure: Articles 74 and 79 were confirmed via full official-page fetches. Articles 81 and 83 were confirmed only via the Service Desk's own indexed search-result snippets, not a full page fetch — a weaker confirmation, flagged in the sources and in Open Questions below. A follow-up verification pass is recommended before treating this article as fully verified to the project's usual standard.

Executive summary

What this means

Who this applies to

Regulatory context

Requirements

What organizations must do

  • Identify the correct market surveillance authority for each system before an issue arises — an organisation with both Annex I sectoral systems and Annex III systems may face genuinely different regulators for different products.
  • Treat any market surveillance evaluation as time-pressured from the outset — 15 working days is the outer limit for corrective action, not a guaranteed grace period.
  • Build Union-wide remediation capability rather than assuming a national finding stays contained to one jurisdiction — the corrective-action duty extends to every instance made available across the Union market.
  • Run an internal audit against Article 83's five formal non-compliance triggers as a distinct, lower-bar compliance check separate from substantive risk assessment.

Common mistakes

Assuming a single "AI regulator" per Member State handles all AI Act enforcement is a common mistake — Article 74's jurisdictional rules split authority by sector, deployer type, and Annex III category.

Treating a market surveillance finding as final once corrective action is taken locally is a second mistake — the cross-notification duty and the Article 81 safeguard procedure mean a national measure can still be escalated or contested after initial local action.

Conflating formal non-compliance under Article 83 with the risk-based procedure under Article 79 is a third mistake — a system can be procedurally non-compliant while presenting no actual risk, and the authority response differs accordingly.

Open questions

  • Full paragraph-by-paragraph confirmation of Articles 81 and 83, and their Digital Omnibus disclaimer status, at the same standard as Articles 74 and 79 — not yet independently retrieved in full.
  • Article 83's specific correction timeframes beyond the five-trigger list — only the Commission's own summary characterisation was confirmed on this pass.
  • How the Article 81 Union safeguard procedure has actually been used in practice since the relevant high-risk provisions began applying — not addressed by any source consulted here.

Frequently asked

FAQ
Did the Digital Omnibus amend Articles 74, 79, 81, or 83?
No disclaimer appeared on any of the four pages checked, though Articles 81 and 83 were only checked via partial snippet retrieval rather than a full page fetch — a weaker confirmation than for Articles 74 and 79.
FAQ
What are Article 83's exact correction timeframes for formal non-compliance?
Not independently retrieved in full on this pass; only the Commission's own summary characterisation was confirmed.
FAQ
How often has the Article 81 Union safeguard procedure actually been invoked?
Not addressed by any source consulted on this pass.

Go deeper

Apply this in practice

AGNUS Academy turns AI governance research into hands-on skills — exercises, case studies, and frameworks built on the research you just read.

Explore the Academy →