In this article11 sections
- What Article 73 Actually Says
- What It Means If You Only Use the System
- When This Applies
- The Legal Definition of "Serious Incident"
- Examples That Would Trigger Action
- Examples That Do Not Trigger Reporting
- Your Response Process as a Deployer
- Building an Internal Incident Template
- The Link to GDPR Incident Reporting
- One Action to Take This Week
- References
AI act serious incident reporting is often explained wrongly. The duty to report under Article 73 sits with the provider of a high-risk AI system, not with the company that uses it. If you deploy a high-risk system, your duty is different: alert the provider and the authority straight away, under Article 26(5). There is no 72-hour rule in the AI Act.
Here is who does what, when it applies, and what your internal process should look like.
What Article 73 Actually Says
Under Article 73 of the EU AI Act, providers of high-risk AI systems must report any serious incident to the market surveillance authority of the Member State where it occurred. The report is due immediately after the provider has established a causal link between the system and the incident, or the reasonable likelihood of one, and at the latest:
-
15 days after becoming aware of the serious incident (the general rule)
-
10 days in the event of a person's death
-
2 days for a widespread infringement or a serious and irreversible disruption of critical infrastructure
A provider may file an incomplete initial report and complete it later.
What It Means If You Only Use the System
Most SMEs are deployers: they use a high-risk system someone else built. Under Article 26(5), a deployer that identifies a serious incident must immediately inform first the provider, and then the importer or distributor and the relevant market surveillance authority. Article 73 applies to the deployer directly only if it cannot reach the provider.
When This Applies
The high-risk rules, including the deployer duties in Article 26, apply from 2 December 2027 for Annex III systems (such as recruitment, credit scoring and emergency triage) and from 2 August 2028 for AI in products covered by Annex I (such as medical devices). The AI Omnibus (Regulation (EU) 2026/1744) set these fixed dates. That leaves time to build a calm, simple process now.
Breaching deployer duties under Article 26 can be fined up to €15 million or 3% of worldwide annual turnover, whichever is higher; for SMEs and start-ups the lower of the two applies (Article 99). National law sets the actual penalties.
In the Netherlands, the draft implementing law (Uitvoeringswet AI-verordening) assigns market surveillance to the AP, the RDI and sector supervisors, such as the IGJ for health and the AFM and DNB for finance. That law is not yet in force, so no Dutch authority has AI Act powers today.
The Legal Definition of "Serious Incident"
Not every AI malfunction counts. The Act defines a serious incident as an incident or malfunction of an AI system that directly or indirectly leads to:
-
Death or serious harm to health of a person
-
A serious and irreversible disruption of critical infrastructure
-
An infringement of Union law obligations that protect fundamental rights
-
Serious harm to property or the environment
Minor errors that cause inconvenience do not qualify. A CV-screening tool that systematically excludes a protected group, with documented discrimination as a result, may well qualify.
Examples That Would Trigger Action
-
Recruitment AI and discriminatory outcomes. Your AI recruitment tool rejects all applicants from one age group because of model bias, and a formal complaint follows. As deployer, you alert the provider and the authority; the provider investigates and reports.
-
Medical AI and patient harm. A clinical decision support tool (a medical device under Annex I) gives a recommendation that leads to delayed treatment and serious injury.
-
Credit-scoring AI and financial exclusion. A credit scoring tool wrongly flags hundreds of people as high-risk, denying them access to credit.
-
Safety component failure. An AI safety component in a lift (a product under Annex I) misses a critical fault, causing an injury.
In each case: a high-risk AI system, serious harm, and a direct or indirect link.
Examples That Do Not Trigger Reporting
-
An AI-generated document contains errors that are caught before use
-
A high-risk system produces an unexpected output but no harm follows
-
A temporary outage with no effect on decisions
-
An employee raises a concern about output quality, but nobody is harmed
Log these anyway. They feed the monitoring duty in Article 26(5), and they help the provider spot problems early.
Your Response Process as a Deployer
Detect and confirm. Identify that a high-risk AI system was involved and that serious harm occurred or is plausible.
Contain. If using the system as instructed may present a risk, Article 26(5) requires you to inform the provider or distributor and the market surveillance authority, and to suspend use.
Notify immediately. Inform the provider first, then the importer or distributor and the market surveillance authority. Keep a timestamped copy of every message.
Support the investigation. The provider must investigate and report under Article 73. Share your logs and facts. Deployers must keep system logs for at least six months (Article 26(6)).
Building an Internal Incident Template
Write the procedure down before you need it. At minimum, capture:
-
Date and time the incident was detected
-
AI system identifier (name, version, provider, intended purpose)
-
Description of what occurred (factual, not interpretive)
-
Persons affected (number, category, nature of harm)
-
Immediate actions taken (system paused, escalation path activated)
-
Who was notified and when (provider, distributor, authority)
Keep the incident log as a living document: every near-miss, complaint and odd output gets a record. It shows that you monitor the system's operation as Article 26(5) requires.
The Link to GDPR Incident Reporting
Many AI incidents also involve personal data, so GDPR Article 33 (72 hours for a personal data breach) may apply alongside the AI Act. The two tracks run in parallel; filing one does not satisfy the other. Your procedure should name who handles each, and your DPO should be involved early.
One Action to Take This Week
If you use an AI system that could be high-risk (recruitment, credit, emergency triage, access to essential services), name the person responsible for incident escalation, and agree with your provider how and where you report incidents to them.
Not sure which of your AI systems are high-risk? Run the free 2-minute compliance check at comply.khairos.ai to see where you stand.
References
Not sure where your company stands?
Answer 15 questions about how your team uses AI. You get a compliance score and the gaps to close first, as a PDF.