Who Enforces the EU AI Act for Medical AI: Notified Bodies, Device Regulators, or AI Authorities?

Why the question has no single answer

Organizations that supply or deploy AI in medical contexts deal with several regulators at once. Medical device rules, the EU AI Act, and national designations all apply, and each assigns a different job to a different body. Treating them as one regulator leads to gaps, such as nobody owning the response to a post-market order.

The AI Act splits enforcement into layers. Notifying authorities oversee the bodies that assess conformity. Market surveillance authorities supervise products after they reach the market. The AI Office, the European Data Protection Supervisor and national competent authorities share wider enforcement, and the Commission states that this enforcement is shared rather than centralized. Enforcement powers of the AI Office and national competent authorities apply from 2 August 2026.

For a compliance leader, the useful question is therefore not who enforces the Act, but which body acts at which stage: before market placement, after it, or across Member States.

Responsibility map by authority role

The table separates the roles the AI Act defines. Which named authority fills each role depends on national designation in every Member State where the product is marketed.

• Notifying authority: Each Member State designates or establishes at least one. It is responsible for the procedures to assess, designate and notify conformity assessment bodies, and to monitor them.
• Notified body: A conformity assessment body that the notifying authority has designated and notified. It performs the third-party assessment step before a product is placed on the market.
• Market surveillance authority: The national authority carrying out activities and measures under Regulation (EU) 2019/1020. It can require corrective action, withdrawal or recall.
• Sectoral authority: The market surveillance authority responsible for the Union harmonisation legislation listed in Annex I. For medical AI, this is the device regulator side of the map.
• AI Office: Part of the European Commission. Exclusively competent for certain system categories and expected to be cooperated with on inspections and enforcement in a Member State.
• European Data Protection Supervisor: Named by the Commission as one of the bodies sharing AI Act enforcement.

Conformity assessment: notified bodies and who watches them

Conformity assessment is the step before market placement. A notified body does the assessment, but it does not choose its own mandate. The notifying authority in each Member State assesses, designates, notifies and then monitors it. A compliance team therefore deals with two different bodies on this front: the notified body that examines the product, and the national authority that decides whether that body is entitled to act.

The Act also sets a special case that does not apply to typical medical AI. Where a high-risk system is intended for use by law enforcement, immigration or asylum authorities, or by Union institutions and bodies, the relevant market surveillance authority acts as the notified body. Medical device suppliers should not assume this arrangement carries over to their products.

Because national designations differ, the notified body and notifying authority for a given device should be recorded per product and per Member State. Keep the date of each designation check, since the Act was amended in 2026.

Market surveillance and corrective powers after placement

Once a product is on the market, supervision moves to market surveillance authorities. Under the AI Act this means the national authority acting under Regulation (EU) 2019/1020, although a Member State may in appropriate circumstances designate another relevant authority. Member States must also ensure coordination with the sectoral market surveillance authorities for Annex I harmonisation legislation. For a medical AI supplier, that coordination is the point where device regulators and AI authorities meet, and it is worth knowing in advance which of them would lead.

The corrective powers are concrete. If a market surveillance authority finds an AI system non-compliant, it must without undue delay require the operator to take corrective action, withdraw the system or recall it. The authority sets the period, which must in any event be within the shorter of 15 working days or the period provided in the relevant Union harmonisation legislation.

That window is short for an organization that has to reconstruct what a system did. The practical consequence is that evidence must already exist, be retrievable, and have a named owner before any order arrives.

The AI Office, the EDPS and when central enforcement applies

The AI Office does not supervise all high-risk AI. It is exclusively competent for defined categories, including systems provided by law enforcement authorities, border management authorities and financial institutions that fall under Article 74(6). Medical AI is not among those examples, so a medical AI portfolio should expect national authorities to be the first point of contact, with the AI Office involved through cooperation. Authorities applying the Act must cooperate actively with the AI Office, including on inspections or enforcement measures in a Member State.

Document which category each product falls into rather than assuming central enforcement. The Digital Omnibus on AI, Regulation (EU) 2026/1744, amended the AI Act and was published in the Official Journal on 24 July 2026. The Commission had proposed to reinforce AI Office powers and reduce governance fragmentation, so governance provisions are worth rechecking against the consolidated text dated 27 July 2026.

Dates that shape planning (as of November 2026)

Timelines moved with the Digital Omnibus. Plan against the amended dates and confirm them in the consolidated text.

• 24 July 2026: Regulation (EU) 2026/1744 published in the Official Journal. It enters into force on the third day after publication.
• 2 August 2026: Enforcement powers of the AI Office and national competent authorities apply.
• 2 December 2027: Fixed deadline reported for stand-alone high-risk AI systems.
• 2 August 2028: AI Act rules for AI embedded in physical products, including medical devices, apply.

Runtime evidence for deployers and providers

Whichever body asks, the underlying question is the same: what was the system allowed to do, and what did it do? A notified body, a device regulator and an AI market surveillance authority may each ask it at a different stage, but all of them need records that are durable and attributable to a specific system and period.

For deployers that use AI tools in clinical, operational or software development workflows, this is where runtime control matters. Trussed enforces policy in the execution path across models, agents, tools and MCP servers, with actions such as allow, redact, block and escalate for human approval, applied to individual interactions and tool calls. That enforcement generates audit-ready evidence mapped to the EU AI Act, which supports deployer-side oversight and record-keeping expectations. Self-managed deployment in the customer cloud, with configurable retention and data residency, lets a regulated enterprise decide where that evidence lives.

This does not replace conformity assessment, which remains the work of notified bodies, and it does not cover model or dataset provenance. It addresses the narrower question of what happened at runtime.