Serious Incident Reporting Deadlines Under the EU AI Act

Article 73 is usually summarised as “fifteen days”. It is three deadlines, not one, and the shortest is two days. Which one applies depends on the category of the incident, and all three run from a moment that is easy to get wrong.

What counts as a serious incident

The clock is only started by a serious incident, which is defined in Article 3(49) of Regulation (EU) 2024/1689 as an incident or malfunctioning of an AI system that directly or indirectly leads to any of four things:

• the death of a person, or serious harm to a person’s health;
• a serious and irreversible disruption of the management or operation of critical infrastructure;
• the infringement of obligations under Union law intended to protect fundamental rights;
• serious harm to property or the environment.

The third limb is the one that makes this provision much broader than a product safety regime, and it is the one teams tend not to have a detection path for. A discriminatory outcome from a hiring or credit system, or a processing operation that breaches data protection law, can be a serious incident under Article 3(49)(c) without anyone being physically harmed. Note also the words “directly or indirectly”: the AI system does not have to be the sole cause.

The three clocks

Article 73(1) requires providers of high-risk AI systems placed on the Union market to report any serious incident to the market surveillance authorities of the Member States where the incident occurred. The deadline then splits:

• Fifteen days — the general rule. Article 73(2): the report is made immediately after the provider establishes a causal link between the AI system and the serious incident, or the reasonable likelihood of such a link, and in any event not later than 15 days after the provider — or, where applicable, the deployer — becomes aware of the incident. The same paragraph says the reporting period is to take account of the severity of the incident, so fifteen days is a ceiling, not an entitlement.
• Ten days — death of a person. Article 73(4): in the event of the death of a person, the report is provided immediately after the provider or deployer has established, or as soon as it suspects, a causal relationship, and not later than 10 days after becoming aware.
• Two days — widespread infringement or critical infrastructure. Article 73(3): in the event of a widespread infringement, or a serious incident within the critical infrastructure limb of Article 3(49)(b), the report is provided immediately and not later than 2 days after becoming aware.

“Widespread infringement” is itself a defined term in Article 3 — broadly, an act or omission contrary to Union law protecting individuals’ interests that has harmed or is likely to harm the collective interests of individuals in at least two Member States other than the one where the act originated, or where the operator is established. A defect that shipped to customers across the Union is far more likely to meet that definition than an isolated failure at one deployer, and the difference between the two-day and the fifteen-day clock is precisely that assessment.

When the clock starts

All three deadlines run from awareness, not from confirmation. Article 73(2) and (4) both refer to the provider or, where applicable, the deployer becoming aware of the serious incident — so a deployer’s knowledge can start the provider’s clock. That single drafting choice has a practical consequence worth designing for: the contractual and technical channel through which deployers tell you about incidents is part of your compliance posture, because a report sitting unread in a shared inbox for eleven days leaves four.

Article 73 anticipates that you will not have a complete picture in time. A provider or deployer may submit an initial, incomplete report followed by a complete one where that is necessary to ensure timely reporting. The correct behaviour under time pressure is therefore to file what is known and supplement it — not to hold the report while the investigation finishes. That is the single most useful operational fact on this page.

What you must do after reporting

Reporting is not the end of the obligation. Following the report, the provider must without delay carry out the necessary investigations into the incident and the system concerned, including a risk assessment of the incident and corrective action, and must cooperate with the competent authorities and, where relevant, the notified body.

There is a preservation duty attached. The provider must not perform any investigation that involves altering the AI system in a way that may affect the subsequent evaluation of the causes of the incident before informing the competent authorities of that action. In engineering terms: the instinct to hot-fix and redeploy is in direct tension with the Regulation. Snapshot the model version, the configuration and the automatically generated logs required by Article 12, tell the authority what you intend to change, and keep the artefacts needed to reconstruct the failure. Article 20 corrective action duties run alongside this, not instead of it — and a post-market monitoring plan that already names its data sources is what makes the investigation tractable inside the deadline.

Where the report concerns the fundamental rights limb, the receiving market surveillance authority is required to inform the national public authorities or bodies designated under Article 77 that supervise fundamental rights obligations. Expect the report to reach a data protection or equality body, not only a product regulator.

Where other reporting regimes take over

Article 73 contains two carve-outs designed to stop double reporting, and both narrow the duty to the fundamental rights limb only:

• Sector regimes with equivalent obligations. For Annex III high-risk systems placed on the market by providers already subject to Union legislative instruments laying down reporting obligations equivalent to those in the AI Act, notification is limited to serious incidents within Article 3(49)(c).
• Medical devices and IVDs. For high-risk systems that are safety components of, or are themselves, devices covered by Regulation (EU) 2017/745 or Regulation (EU) 2017/746, notification is likewise limited to the Article 3(49)(c) category, and is made to the national competent authority chosen for that purpose by the Member State where the incident occurred. The vigilance obligations under the device regulations continue to apply for everything else — see how the MDR and the AI Act interact.

The general reporting obligations in Article 73 apply along with the rest of the high-risk regime, and that date has moved: from 2 August 2026 to 2 December 2027 for stand-alone Annex III systems, by Regulation (EU) 2026/1744, which was adopted on 8 July 2026, published in the Official Journal on 24 July 2026 and entered into force on 27 July 2026. For high-risk AI embedded in the Annex I regulated products it moved from 2 August 2027 to 2 August 2028. The clocks themselves — 15 days, 10 days, 2 days — are unaffected; what moved is the day from which they start binding. Providers of general-purpose AI models with systemic risk have had a separate and differently worded duty since 2 August 2025 under Article 55, which requires tracking, documenting and reporting serious incidents to the AI Office and, as appropriate, national competent authorities without undue delay — no fixed day count. Do not read the Article 73 clocks onto Article 55, or the other way round.