Reporting duties after an AI agent incident: AI Act, CRA, NIS 2 and GDPR

Side by side

	AI Act	Cyber Resilience Act	NIS 2 Directive	GDPR
Trigger	Serious incident of a high-risk AI system: death or serious harm to health, serious and irreversible disruption of critical infrastructure, infringement of fundamental-rights obligations, serious harm to property or the environment	Severe incident affecting the security of a product with digital elements, or an actively exploited vulnerability	Significant incident affecting the services of an essential or important entity	Personal data breach
Who reports	Provider; the deployer informs the provider first, then importer or distributor and authorities	Manufacturer	The entity	Controller
To whom	Market surveillance authority of the Member State where it occurred	Single reporting platform: CSIRT designated as coordinator and ENISA; impacted users are informed	National CSIRT or competent authority	Supervisory authority; data subjects where the risk is high
Deadline	Not later than 15 days; 10 days after a death; 2 days for critical infrastructure or a widespread infringement	Early warning 24 h, notification 72 h, final report one month after the notification	Early warning 24 h, notification 72 h, final report one month later	Without undue delay, where feasible within 72 h
Applies	With the high-risk obligations: December 2, 2027 (Annex III), August 2, 2028 (Annex I Section A)	Since September 11, 2026	Through national transposition	Since May 25, 2018

What the agent changes — and what it doesn’t

An incident is not reportable because an agent was involved, and not exempt because “only software” acted. What counts is the consequence and the company’s role: provider or deployer of an AI system, manufacturer of a product, essential or important entity, controller of personal data. One event can fall under several rules — a manipulated assistant in a connected product that leaks customer data and changes device configurations can be a CRA severe incident and a GDPR breach at once.

Per system type

• IT systems: Data leaks through an agent are the typical GDPR case; for entities under NIS 2, outages of essential services add their duties.
• IoT products: The agent in a product, its app or its remote data processing is part of the product; manipulations that affect its security functions can be CRA severe incidents.
• OT: Operators of critical infrastructure may be under NIS 2; an AI system that is a safety component in the management of critical infrastructure is high-risk under Annex III of the AI Act.
• Machines: For machinery, which sits in Annex I Section B since Regulation (EU) 2026/1744, AI functions are assessed under the Machinery Regulation from January 20, 2027; its recording duties (Annex III, 1.1.9 and 1.2.1) keep the evidence an investigation needs. Check which reporting route applies before relying on Article 73 AI Act.

The evidence behind the report

Reports rest on the secured agent trace, the logs required for high-risk systems (record-keeping and logs) and the triage record. Incomplete initial reports are allowed under the AI Act and expected under the CRA and NIS 2; waiting for full certainty is not.

This card names possible duties; it is no legal advice, and naming a rule here never establishes compliance. The whole picture is on the pathway map on AI agents crossing the line.