Skip to main content
Praxikon

Reporting a serious incident with high-risk AI: what Article 73 of the AI Act asks of a municipality

··21 min read

When a municipality identifies a serious incident involving a high-risk AI system it uses, it immediately informs first the provider, and then the importer or distributor and the market surveillance authority (Article 26(5) AI Act). The formal report is made by the provider: to the market surveillance authority of the Member State where the incident occurred, immediately after it has established a causal link (or the reasonable likelihood of one), and no later than 15 days after it or, where applicable, the deployer became aware of the incident. In case of death the limit is 10 days, and for a widespread infringement or a serious and irreversible disruption of critical infrastructure it is 2 days (Article 73(1) to (4)). If the municipality cannot reach the provider, Article 73 applies to the municipality itself. Because the clock can start running when the municipality discovers the incident, the reporting chain belongs in the contract with the supplier and in an internal process of its own.

Suppose the Work and Income department uses a supplier's AI system that assesses applications for social assistance for the risk of irregularity and gives the caseworker a recommendation. After an update, a team leader notices that applications from one neighbourhood are flagged as high risk remarkably often, and that caseworkers who follow those recommendations order extra checks and stop payments. Is that a serious incident? Who has to do what, and how fast? This piece works through the questions.

In short

A serious incident is an incident or malfunctioning of an AI system that directly or indirectly leads to death or serious harm to health, a serious and irreversible disruption of critical infrastructure, an infringement of obligations under Union law intended to protect fundamental rights, or serious harm to property or the environment (Article 3(49)). The provider reports it to the market surveillance authority (Article 73(1)), within 15, 10 or 2 days (paragraphs 2 to 4), if necessary first with an incomplete report (paragraph 5), then investigates it, and does not alter the system in a way that may affect the evaluation of the cause without first informing the authorities (paragraph 6). The municipality as deployer monitors the system and, in the event of a serious incident, immediately informs first the provider, then the importer or distributor and the market surveillance authority (Article 26(5)). The provider feeds those signals into its post-market monitoring (Article 72). For Annex III systems these obligations bite from 2 December 2027; for systems already in use earlier, the transitional rule of Article 111(2) applies.

This is the chapter on Article 73 itself. If the incident also involves leaked personal data, the GDPR route with its 72 hours runs alongside it; that is worked out in data breach in an AI system: notifying the authority and the people affected. What the Commission put forward in 2025 as draft guidance and a reporting template is covered in our piece on the incident reporting consultation.

What is a serious incident?

Article 3(49) defines a serious incident as "an incident or malfunctioning of an AI system" that directly or indirectly leads to one of four outcomes:

PointOutcomeDeadline for the provider
aThe death of a person, or serious harm to a person's healthNo later than 15 days; in case of death no later than 10 days (paragraph 4)
bA serious and irreversible disruption of the management or operation of critical infrastructureNo later than 2 days (paragraph 3)
cThe infringement of obligations under Union law intended to protect fundamental rightsNo later than 15 days; no later than 2 days if it is also a widespread infringement (paragraph 3)
dSerious harm to property or the environmentNo later than 15 days

Three things stand out. First, "indirectly" is enough. The AI system does not have to cause the harm itself; a recommendation that an employee follows and that leads to the harm can suffice. Second, point (c) matters most for public bodies and banks. A system that systematically disadvantages a group of people can lead to an infringement of obligations that protect fundamental rights, such as the prohibition of discrimination or obligations under the GDPR, without anyone being physically hurt. Third, a widespread infringement is a defined term (Article 3(61)): an act or omission contrary to Union law protecting the interest of individuals which (a) has harmed or is likely to harm the collective interests of individuals residing in at least two Member States other than the Member State in which the act or omission originated or took place, in which the provider or its authorised representative is located or established, or, where the deployer committed the infringement, in which the deployer is established; or (b) has caused, causes or is likely to cause harm to the collective interests of individuals, has common features such as the same unlawful practice or the same interest being infringed, and is occurring concurrently, committed by the same operator, in at least three Member States. For a municipality serving only its own residents, that will rarely apply.

Critical infrastructure has the meaning of Article 2, point (4), of Directive (EU) 2022/2557 (Article 3(62)). A municipality that uses AI in managing, for example, water or traffic should check that definition separately.

A serious incident is not the same as a risk. Article 26(5) has two tracks. Where the municipality has reason to consider that use in accordance with the instructions may result in a risk within the meaning of Article 79(1) (a risk to health, safety or fundamental rights), it informs the provider or distributor and the market surveillance authority without undue delay, and suspends use. Suspending is not optional. A serious incident is the next step: the harm has occurred, or the system has directly or indirectly led to it. Then the immediate reporting chain of Article 26(5) and the deadlines of Article 73 apply. In practice the first signal often comes in through the risk track, so decide in advance who at the municipality determines which track a report is in.

Who reports what to whom

PartyObligationTo whomDeadlineLegal basis
Deployer (the municipality)Serious incident identified: informFirst the provider, then the importer or distributor and the market surveillance authorityImmediatelyArticle 26(5)
DeployerProvider unreachable: report itselfMarket surveillance authorityDeadlines of Article 73, applied mutatis mutandisArticle 26(5)
ProviderReport the serious incidentMarket surveillance authority of the Member State where it occurredImmediately after establishing a causal link, no later than 15, 10 or 2 days after awarenessArticle 73(1) to (4)
ProviderInvestigation, risk assessment and corrective actionCooperation with the competent authoritiesWithout delay after the reportArticle 73(6)
Market surveillance authorityFor an incident under point (c): inform other authoritiesNational public authorities protecting fundamental rights (Article 77(1))Upon receiptArticle 73(7)
Market surveillance authorityTake appropriate measuresUnder Article 19 of Regulation (EU) 2019/1020Within seven days of receiptArticle 73(8)

The clock may already run from your discovery. Paragraphs 2 to 4 also refer to when 'where applicable' the deployer becomes aware. Whether that starts the provider's clock at your discovery is unsettled; to be safe, assume it does. If the municipality discovers the incident on Monday and only informs the supplier a week later, much of the provider's deadline may already be gone. That is why Article 26(5) says "immediately". For a death the bar is higher still: the report follows immediately after the provider or the deployer has established, or as soon as it suspects, a causal relationship (paragraph 4).

Report first, know everything later. The reporting period takes account of the severity of the incident (paragraph 2, second subparagraph). Where necessary to ensure timely reporting, the provider or the deployer may submit an initial incomplete report, followed by a complete one (paragraph 5). Waiting until the cause is fully known is therefore no reason to let the deadline pass.

Change nothing without saying so. After the report, the provider investigates the incident, including a risk assessment and corrective action. It cooperates with the competent authorities and does not perform any investigation that involves altering the system in a way that may affect the subsequent evaluation of the causes, without informing the authorities beforehand (paragraph 6). For the municipality this means that a quick "hotfix" by the supplier can wipe away the evidence. Agree that the supplier will not do this without consultation, and make sure the logs of the incident are not deleted automatically; see logs of high-risk AI systems.

Exception for law enforcement. The obligation in Article 26(5) does not cover sensitive operational data of deployers that are law enforcement authorities.

Post-market monitoring: Article 72

Reporting is the end point; monitoring is the source. The provider must establish and document a post-market monitoring system in a manner proportionate to the nature of the AI technology and the risks (Article 72(1)). That system actively and systematically collects, documents and analyses relevant data on the performance of the system throughout its lifetime, including data that deployers may provide, so that the provider can evaluate continuous compliance with the requirements (paragraph 2). The system is based on a plan that forms part of the technical documentation; the Commission adopts guidelines including a template for that plan by 2 September 2027 (paragraph 3, as amended by Regulation (EU) 2026/1744).

The two sides fit together. The municipality monitors operation on the basis of the instructions for use and informs the provider "in accordance with Article 72" (Article 26(5)). What the municipality sees in its own monitoring (skewed outcomes per group, a rise in objections, caseworkers following or overriding the recommendation remarkably often) is exactly the input the provider needs to spot incidents early. So agree which signals the municipality passes on, in what form and how often. If that data contains personal data, the disclosure is a processing operation of its own that needs a legal basis; aggregated figures are often enough.

If the provider finds that the system does not conform to the Regulation, it immediately takes corrective action and informs distributors and, where applicable, deployers (Article 20(1)). That information, too, must reach the right person at the municipality.

What the municipality arranges in the contract

The AI Act places the reporting obligation on the provider, but the deadline can start running at the municipality. Anyone who leaves the contract unchanged will only find out when things go wrong. In our reading the agreement with the supplier should cover at least these points:

  1. A fixed point of contact, both ways. A contact point at the supplier that can be reached outside office hours, so that the municipality can "immediately inform first the provider", and a contact point at the municipality for the supplier's notices about corrective action and risks.
  2. A deadline in hours for feedback. When does the supplier confirm receipt, when does it say whether it sees a causal link, and when does it file its report with the market surveillance authority? Ask for a copy or summary of that report. The draft guidance reads 'immediately' as within 24 hours, and the provider as unreachable if it does not answer within 24 hours (points 47 and 48).
  3. Information the supplier needs. Agree which data the municipality provides (time of discovery, model version involved, examples, number of people affected) and in what form.
  4. Access to logs and a retention hold. The municipality must be able to view and export the logs under its control, and in an incident the automatic deletion on both sides must be suspendable.
  5. No changes without consultation. An agreement that the supplier will not alter the system during the investigation in a way that may affect the evaluation of the cause, without first informing the municipality and the authorities (Article 73(6)).
  6. Monitoring data. Which data the municipality shares for the purposes of Article 72, how often, and whether it can be aggregated.
  7. Roles under the GDPR. If the incident also involves personal data, the data processing agreement applies with its own notification deadline. Agree that a single signal from the supplier triggers both routes.

What the municipality arranges in its own process

  • Who identifies? Article 26(5) speaks of a deployer that "identifies" a serious incident. Designate a function that makes that call, with a deputy, and decide that in case of doubt you report first and assess later. An initial incomplete report is allowed (Article 73(5)).
  • One triage for all reporting duties. The same event can be a serious incident, a personal data breach and, for organisations covered by the Dutch Cybersecurity Act (Cyberbeveiligingswet, implementing NIS2), a significant incident at the same time. Let one team triage and keep the clock per regime; the GDPR and NIS2 routes are set out in the data breach piece.
  • Human oversight as a signal. Caseworkers who assess the system's recommendations are often the first to notice something. Give them a simple route to report anomalies; the human oversight of Article 26(2) only becomes useful when signals actually arrive somewhere.
  • Being able to suspend. Decide who may suspend use in case of a risk (Article 26(5)), and how the service continues in the meantime.
  • Record. Keep your own register of signals, assessments and reports, with timestamps. That way you can later show when the municipality became aware of the incident and when it informed the provider.
  • Reporting when the provider is silent. If the municipality cannot reach the provider, Article 73 applies mutatis mutandis and it reports to the market surveillance authority itself.

The municipality in the example

Back to Work and Income. Assume the system was put into service after 2 December 2027 (see 'From when?'). A system used by or on behalf of public authorities to evaluate whether people are eligible for essential public assistance benefits, or to grant, reduce, revoke or reclaim them, is listed in Annex III, point 5(a), and is then a high-risk system (unless the exception in Article 6(3) applies, which never applies where the system performs profiling).

The team leader's first signal is a reason to consider that the system may present a risk to fundamental rights. That is the risk track: the municipality informs the supplier and the market surveillance authority without undue delay, and suspends use. If the analysis shows that payments were stopped on the basis of recommendations that systematically disadvantaged people from one neighbourhood, that can lead to an infringement of obligations protecting fundamental rights, such as the prohibition of discrimination. In our reading it is then a serious incident under Article 3(49)(c), even though a caseworker took the decision: the recommendation led to it indirectly. The municipality immediately informs the supplier, and the supplier reports no later than 15 days after it, and possibly already after the municipality, became aware of it. The market surveillance authority then passes the report on to the national authorities protecting fundamental rights (Article 73(7)). For the residents affected, the right to an explanation under Article 86 and the municipality's own objection procedure also come into play.

To whom does a Dutch municipality report? That is not settled yet. The Netherlands has not yet designated a market surveillance authority for the AI Act. According to the Planning Letter on Digital Affairs (Parliamentary Paper 26643, no. 1554), the bill for the Dutch AI Regulation Implementation Act is scheduled to be submitted to the House of Representatives in the second quarter of 2027. That is the date of submission, not the date the law applies. So record "which authority" as an open point in the process and update it once the designation is made.

Example 2: the bank and credit scoring

A bank that buys an AI system from a supplier to evaluate the creditworthiness of individuals uses a high-risk system under Annex III, point 5(b) (except systems used to detect financial fraud). Here too, assume the system was put into service after 2 December 2027. Three particulars apply to the bank.

Monitoring through governance. For deployers that are financial institutions subject to internal governance requirements under Union financial services law, the monitoring obligation of Article 26(5) is deemed fulfilled by complying with those governance rules (Article 26(5), second subparagraph). In our reading that does not remove the duty to immediately inform the provider of a serious incident; the safe course is to set up that reporting route separately.

Equivalent reporting obligations. For Annex III systems placed on the market or put into service by providers already subject to Union instruments with equivalent reporting obligations, reporting under the AI Act is limited to infringements of fundamental rights, point (c) (Article 73(9)). If the bank develops the system itself, it is the provider and the question may arise for it. The Regulation does not list them. The Commission's 2025 draft guidance treats DORA as equivalent (points 56 to 61), but it is not adopted or binding; so substantiate the limitation yourself. A similar limitation applies to AI in medical devices under Regulations (EU) 2017/745 and 2017/746, with reporting to the authority designated for that purpose (paragraph 10).

Integrating the monitoring plan. A financial institution that places an Annex III, point 5 system on the market as a provider may integrate the elements of Article 72 into systems and plans it already has under financial services law, provided this achieves an equivalent level of protection (Article 72(4), second subparagraph).

An incident under point (c) at a bank could be, for example, a model that after an update systematically rejects applicants with a particular characteristic. The same chain arrangements as at the municipality are then needed: a fixed point of contact at the supplier, a deadline in hours, and a retention hold on the logs. In addition, the same event may be an ICT-related incident under DORA or a personal data breach under the GDPR, with their own deadlines and recipients.

From when?

Articles 72 and 73 are in Chapter IX and, under Article 113, have applied since 2 August 2026. The obligations, however, rest on providers of high-risk AI systems, and the rules that determine which system is high-risk, and the deployer obligations in Article 26, are in Chapter III, Sections 1 to 3. Those apply from 2 December 2027 for Annex III systems and from 2 August 2028 for Annex I systems (Article 113, third paragraph, point (c), as amended by Regulation (EU) 2026/1744). In practice the reporting obligation therefore only has something to bite on from those dates. For high-risk systems already placed on the market or put into service before that date, the Regulation applies only if, as from that date, they undergo significant changes in their design, and systems intended to be used by public authorities must comply by 2 August 2030 at the latest (Article 111(2)). Regulation (EU) 2026/1744 did not change the deadlines in Article 73 itself.

The Commission's guidance. Article 73(7) requires the Commission to develop dedicated guidance to facilitate compliance with the reporting obligation, to be issued by 2 August 2025. In September 2025 the Commission put draft guidance and a reporting template out for consultation. On its overview page on guidelines the Commission lists a reporting template for serious incidents as published, but the link there does not work, and on 17 November 2026 we found no adopted version of the guidance or the template. Until there is one, the text of Article 73 is the yardstick; we will update this piece once the final guidance appears.

What to do now

  1. Take stock of the AI systems you use that fall, or could fall, under Annex III, and who the provider is for each.
  2. Amend the contracts with the seven points above, starting with the point of contact and the deadline in hours.
  3. Designate a function that identifies a serious incident and informs the provider, with a deputy.
  4. Build one triage for the AI Act, the GDPR and, where relevant, NIS2 or DORA.
  5. Arrange the logs: access, export and a retention hold for incidents.
  6. Rehearse one scenario per system, with the question: when did we become aware, and when did the provider know?

Unsure whether a system is high-risk and which obligations then apply? The case check lets you describe the use in a few questions and shows which AI Act and GDPR obligations come into view.

Frequently asked questions about reporting serious incidents

Sources

European Commission, AI Act Service Desk: Article 73: Reporting of serious incidents (accessed 17 November 2026)
European Commission: AI Act: draft guidance and reporting template on serious AI incidents (consultation) (26 September 2025, accessed 17 November 2026)
Dutch State Secretary for Economic Affairs: Planningsbrief Digitale Zaken, Kamerstuk 26643, nr. 1554 (letter of 31 August 2026, published 14 September 2026, accessed 17 November 2026)
European Commission: Draft guidance Article 73 AI Act: incident reporting (high-risk AI systems), points 47, 48 and 56 to 61 (26 September 2025 (draft), accessed 17 November 2026)
European Commission: Supporting the implementation of the AI Act with clear guidelines (accessed 17 November 2026)

Newsletter

Every Tuesday, the AI Act week ahead in 5 minutes

A practical briefing on deadlines, new guidance and enforcement, so you know what matters this week. No spam and you can unsubscribe in one click.

Practical and short · No spam · One-click unsubscribe

Written by

Zahed Ashkara, freelance AI and privacy consultant and legal adviser

Zahed Ashkara

Freelance AI and privacy consultant and legal adviser · AI governance & safety

Zahed Ashkara is a freelance AI and privacy consultant and legal adviser for the GDPR and the EU AI Act. Organisations hire him as AI governance consultant, privacy counsel, external DPO or AI governance lead, have him carry out a DPIA, FRIA or AI Act classification, or book an in-company training. He focuses on AI governance and safety, including AI agents. He works exclusively as a freelancer, through Embed AI (trade name Praxikon). He is a Certified AI Compliance Officer (CAICO) and has been a member of the NEN standards committee Artificial Intelligence & Big Data since July 2026.

Follow on LinkedInView expert profile

Use these sources in your AI assistant too: Praxikon in Claude and other assistants

Real-World Stories

Discover how organizations apply this in practice

Related articles

Logs of high-risk AI systems: how long to keep them under the EU AI Act, and does the GDPR allow it?

From 2 December 2027 (for systems already in use, by 2 August 2030 at the latest), a municipality using a high-risk AI system keeps the automatically generated logs under its control for a period appropriate to the intended purpose, of at least six months (Article 26(6) AI Act). In principle the GDPR does not prohibit this, but it does decide how much longer is allowed and how the logs must be secured. What Articles 12, 19 and 26 require, what the logs must record, and how to justify the retention period, with a municipality and a bank as examples.

EU AI ActLoggingRecord-keeping

Accuracy, robustness and cybersecurity of high-risk AI: what Article 15 of the AI Act asks of providers and municipalities

A high-risk AI system must achieve an appropriate level of accuracy, robustness and cybersecurity and keep it up throughout its lifecycle (Article 15 AI Act). That requirement sits with the provider, but the municipality using the system must be able to read the accuracy figures in the instructions for use, use the system in line with those instructions and monitor its operation. What paragraphs 1 to 5 require, how they connect with Articles 9, 10, 26 and 42, the Cyber Resilience Act and the GDPR, and what a municipality puts in its tender, its contract and its own process, with a bank as a second example.

EU AI ActArticle 15Accuracy

The right to explanation under Article 86 of the EU AI Act: what a municipality must explain

Where a municipality takes a decision about a resident on the basis of the output of a high-risk AI system listed in Annex III, that resident can ask for clear and meaningful explanations of the role of the AI system and the main elements of the decision (Article 86 AI Act). Who holds the right, when it applies, what the explanation must contain, how it relates to Article 22 and Article 15(1)(h) GDPR, and what to record now.

EU AI ActArticle 86Right to explanation