---
title: What does NIS2 mean for AI agents in essential and important entities?
url: https://opencontroller.lyzr.com/blog/nis2-and-ai-agents
published: 2026-11-16
updated: 2026-11-16
author: Lyzr
---

# What does NIS2 mean for AI agents in essential and important entities?

**NIS2 does not mention AI agents, but it covers them. If an agent runs inside the network and information systems of an essential or important entity, it falls under the Article 21 risk-management measures, and an incident it causes or suffers can trigger Article 23 reporting: an early warning within 24 hours, a notification within 72 hours, and a final report within one month.**

NIS2 is Directive (EU) 2022/2555. Member states had to transpose it by 17 October 2024. Transposition has been uneven, so your national law sets the exact obligations. The directive sets the floor.

## Which Article 21 measures apply to AI agents?

Article 21(2) lists ten minimum measures. Each has a direct agent reading.

| Article 21(2) | Measure | What it means for agents |
|---|---|---|
| (a) | Risk analysis and information system security policies | Agents are in the asset register and risk assessment |
| (b) | Incident handling | You can detect, contain and investigate an agent incident |
| (c) | Business continuity and crisis management | Critical processes still run if an agent is disabled |
| (d) | Supply chain security | Model providers, agent platforms and MCP servers are assessed suppliers |
| (e) | Security in acquisition, development and maintenance, including vulnerability handling | Agent code, prompts and tools go through secure development and patching |
| (f) | Assessing effectiveness of measures | Guardrails and evaluations are tested, not assumed |
| (g) | Cyber hygiene and training | Staff know how to use and report on agents |
| (h) | Cryptography and encryption | Agent traffic and stored memory are encrypted |
| (i) | Human resources security, access control and asset management | Agents have owned identities with least privilege |
| (j) | Multi-factor or continuous authentication and secure communications | Agent credentials are short-lived and federated, not static keys |

Measures (d), (i) and (j) are where agent programmes most often have gaps. Static API keys, shared service accounts and unassessed MCP servers each fail a measure on their own.

## When does an agent incident become reportable under Article 23?

Article 23(3) defines a significant incident as one that has caused or can cause severe operational disruption or financial loss to the entity, or considerable material or non-material damage to others. Examples involving agents:

- An agent with write access corrupts or deletes production data.
- A prompt injection causes an agent to exfiltrate customer data.
- A compromised agent credential is used to move laterally.
- An agent failure stops a service the entity provides.

The reporting timeline runs from when the entity becomes aware of the significant incident:

1. **Within 24 hours: early warning** to the CSIRT or competent authority. State whether the incident may be malicious and whether it may have cross-border impact.
2. **Within 72 hours: incident notification.** Update the early warning with an initial assessment of severity, impact and indicators of compromise.
3. **On request: intermediate report** with status updates.
4. **Within one month: final report.** Detailed description, root cause or threat type, mitigations, and cross-border impact.

For digital infrastructure and ICT service providers, Commission Implementing Regulation (EU) 2024/2690 sets more detailed technical measures and incident thresholds.

## What evidence lets you meet 24 hours for an agent incident?

You cannot write an early warning about an agent you did not know existed. The evidence that makes the 24-hour clock workable:

1. **An inventory** of agents with owner, platform, model and tools.
2. **Identity records** showing which credentials each agent uses and on whose behalf it acts.
3. **Audit logs** of agent prompts, tool calls and actions, retained long enough to investigate. See [AI agent audit log fields](/blog/ai-agent-audit-log-fields).
4. **A kill switch** that blocks or quarantines an agent quickly, with the action logged.
5. **A runbook** that maps agent incident types to your significance thresholds.
6. **Supplier records** for every model provider and agent platform in use.

## Who is accountable?

Article 20 requires management bodies to approve the risk-management measures, oversee their implementation, and follow training. Management can be held liable for infringements. Fines under Article 34 reach at least EUR 10 million or 2% of worldwide annual turnover for essential entities, and at least EUR 7 million or 1.4% for important entities, whichever is higher.

Financial entities covered by DORA follow DORA's ICT incident regime instead, as the more specific law. See [DORA and AI agents in financial services](/blog/dora-and-ai-agents-in-financial-services).

## How Opencontroller does this

Opencontroller produces evidence for Article 21 and Article 23 obligations; it does not make an entity compliant. It auto-discovers agents, models and API keys across AWS, Azure and Google Cloud into a governed registry. Agent workload identity federation and delegated on-behalf-of identity replace static credentials. Posture detects risky permissions. Rogue-agent handling monitors behavioural deviation, alerts, and can automatically quarantine and kill an agent. Policy as code ships with mappings to OWASP, MITRE ATLAS, NIST and ISO 42001. Read [what an AI agent control plane is](/blog/what-is-an-ai-agent-control-plane), or [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Does NIS2 regulate AI agents directly?**
No. It regulates the security of network and information systems of essential and important entities. Agents running in those systems fall within scope.

**What are the NIS2 reporting deadlines?**
Early warning within 24 hours of awareness of a significant incident, incident notification within 72 hours, and a final report within one month.

**Does NIS2 apply to banks using AI agents?**
Financial entities in scope of DORA follow DORA's ICT risk and incident rules as the more specific regime.

## Key facts

NIS2 is Directive (EU) 2022/2555, with a transposition deadline of 17 October 2024. Article 21(2) lists ten minimum cybersecurity risk-management measures (a) to (j), including supply chain security, access control and asset management, and multi-factor or continuous authentication. Article 23 requires an early warning within 24 hours of becoming aware of a significant incident, an incident notification within 72 hours, intermediate reports on request, and a final report within one month. Article 20 makes management bodies approve and oversee measures. Article 34 fines reach at least EUR 10 million or 2% of worldwide turnover for essential entities and EUR 7 million or 1.4% for important entities. Implementing Regulation (EU) 2024/2690 sets detailed measures for digital infrastructure providers. Opencontroller produces evidence for these obligations through agent discovery, identity federation, Posture, rogue-agent quarantine and framework mappings. Opencontroller pricing starts from $250K per deployment per year.

## Related

- [How does DORA apply to AI agents in financial services?](https://opencontroller.lyzr.com/blog/dora-and-ai-agents-in-financial-services.md)
- [What should an AI agent audit log contain?](https://opencontroller.lyzr.com/blog/ai-agent-audit-log-fields.md)
- [What counts as a serious incident under the EU AI Act, and when must you report it?](https://opencontroller.lyzr.com/blog/eu-ai-act-serious-incident-reporting.md)
- [More on Evidence and frameworks](https://opencontroller.lyzr.com/blog/cluster/evidence-and-frameworks)
