On 7 July 2026, the European Commission published communication COM(2026) 577 final, entitled "EU action plan on cybersecurity and artificial intelligence". This text does not create a new directly enforceable obligation. It schedules nine concrete actions, several with deliverables due before the end of the third quarter of 2026, that is, in the coming weeks. For CISOs of entities subject to NIS2, ignoring this timetable means suffering regulatory references rather than anticipating them.
A bridge between two texts that did not speak to each other
NIS2 requires the management of risks linked to information systems. The EU AI Act requires the governance of AI systems. These two texts were designed almost independently. An AI system deployed in a critical process of an essential entity falls within the scope of both at the same time. GRC teams that handle them separately multiply audits, registers and assessments without producing coherence.
Plan COM(2026) 577 final explicitly serves as a bridge. It tasks ENISA, Member States and sectoral authorities with coordinating their approach across nine actions. The logic is not to add a third text: it is to articulate what already exists. For you, this means that forthcoming ENISA publications will refer to both NIS2 and the EU AI Act. Entities that have not yet aligned their two compliance programmes will have to do so under audit pressure.
What arrives before October 2026
The most immediate action for NIS2 entities is action 4 of the plan: ENISA must publish guidelines on the secure use of AI in the third quarter of 2026. This quarter ends on 30 September. These guidelines will not be an academic document: they will define what "appropriate technical measures" means when the system concerned integrates AI.
Why does this matter now? Because national supervisory authorities use ENISA publications as an assessment benchmark in NIS2 audits. Article 21 of the NIS2 Directive requires measures proportionate to risk, without detailing them. The ENISA guidelines fill that gap. When they are published, they quickly become the operational definition that your inspector will cite.
NIS2 entities that have not yet mapped their AI systems will have to do so in a hurry once the guidelines are published, whereas those that have prepared will be able to respond immediately to an audit questionnaire.
The figures that justify the urgency of the plan
The plan relies on two data points that security teams recognise in the field. First, according to ENISA's Threat Landscape 2025 report, 80% of the phishing campaigns analysed contained AI-generated or AI-personalised elements, compared with less than 20% two years earlier. The AI-related attack surface is no longer a projection: it is already the daily reality of SOC teams.
Second, the financial envelope: EUR 200 million is committed to defensive AI research until the end of the current multiannual financial framework (2027). This amount directly funds the plan's actions, including the test platform and the open source programme. It is a signal that these deliverables carry political priority, not merely an ambition stated in a document.
The test platform: a resource to build into your planning
Action 2 of the plan provides for the establishment, with the Joint Research Centre (JRC), of a secure test platform allowing organisations in the energy, transport, health, finance and public administration sectors to test AI deployments in a simulated environment. It is expected in the fourth quarter of 2026.
In concrete terms, a CISO in the health sector will be able, from the first quarter of 2027, to submit AI-assisted attack scenarios to this platform without exposing production systems. This is the type of tool that teams lacked when they wanted to demonstrate the robustness of their controls against AI threats, not just documentary compliance.
If your entity falls within one of the five covered sectors, including this platform in your 2027 security roadmap is a decision you can take now, before the tool is available.
The open source angle: what the plan brings to your SBOM teams
Action 6 of the plan provides, in the fourth quarter of 2026, for a mapping of critical open source components used in AI systems in essential sectors. This mapping will be followed by a "Critical Open Source Resilience" programme bringing together public and private experts to strengthen these components.
For teams managing SBOMs under NIS2 or the Cyber Resilience Act, this mapping will be a reference: it will identify the AI libraries most present in European critical infrastructure, those whose vulnerability would have the greatest impact. Cross-referencing it with your SBOM inventory will make it possible to prioritise updates and document your prioritisation process during an audit.
Three adjustments to make now
Map the AI systems within your NIS2 scope. The ENISA Q3 2026 guidelines will apply to AI systems already deployed. If you do not yet know which are present in your critical processes, you will not be able to assess your gap against future requirements. The inventory does not need to be exhaustive before October: the essential point is to cover the AI systems in the functions classified as essential in your NIS2 risk analysis.
Check that your risk management policy explicitly covers AI systems. Article 21 of NIS2: your risk management measures must be proportionate. A decision-making AI system in a critical process must appear in your risk register, with its dedicated controls, owners and monitoring criteria. The article on NIS2 and the risk vectors of generative AI details the angles your security policies must address.
Anticipate the convergence of NIS2 and the EU AI Act. The plan explicitly encourages alignment between the two frameworks for high-risk AI systems. Entities that maintain two separate compliance programmes will, as authorities coordinate their audits, generate costly redundancies. Identifying common requirements now reduces future workload.
What this plan does not do
It would be inaccurate to present this plan as a third binding regulation. It is not. It does not replace NIS2, the EU AI Act or the Cyber Resilience Act. It imposes no compliance deadlines, no fines and no mandatory notification. Its binding force is indirect: it operates through the deliverables it commands from ENISA and Member States. When ENISA publishes its guidelines, these become the definition that supervisory authorities use. That is when the plan produces concrete effects on your compliance programme.
The timetable is tight. Q3 2026 ends in five weeks. Organisations that have adapted their risk registers and security policies before this publication will be in a position to respond to the first requests from their supervisory authorities. Others will have to adjust under pressure, at the very moment when NIS2 incident notification deadlines continue to apply without tolerance.