The document is more than another policy statement on artificial intelligence. Its central message is that AI is becoming a strategic cybersecurity capability. Organisations will use it to detect threats, accelerate response, improve vulnerability management and strengthen resilience. Attackers will use it for the same reason: speed, scale and automation.
For European decision-makers, the most important point is not the technology itself. It is dependency. The Commission recognises that many frontier AI capabilities are developed outside the EU, while access to them is often controlled by foreign providers through opaque or provider-specific processes. In sensitive cybersecurity use cases, this creates a strategic problem. Europe may need advanced AI to defend its own critical infrastructure, but the most powerful capabilities may not be fully under European control.
The action plan tries to answer this problem through three main lines of work: making advanced AI safer before deployment, giving trusted European actors clearer routes to access powerful AI capabilities, and helping the cyber ecosystem adapt to AI-powered attacks.
A key proposal is the creation of an EU evaluation capacity for AI models, including cybersecurity evaluation. This would support the Commission’s AI Act oversight and create a European alternative for independent assessment of model capabilities and risk mitigations. In practical terms, the EU wants to know earlier whether powerful models could be misused in cyber operations, and whether providers have taken credible steps to reduce that risk.
The plan also introduces the idea of a European Blueprint for structured access to advanced AI capabilities for cybersecurity purposes. This is one of the most significant parts of the document. Today, access to the most capable AI models may depend on private decisions by providers, including staged releases or trusted-access programmes. The Commission wants clearer criteria for which European organisations should be able to access advanced AI capabilities, including EU institutions, Member State authorities, critical infrastructure operators, cybersecurity providers and researchers.
This is a strong idea. If implemented well, it could reduce uncertainty for European organisations that need advanced AI for legitimate defence purposes. It could also give providers a reference framework for granting access without increasing misuse risks. However, the blueprint will not create new obligations for AI providers. That limits its immediate force. It may become influential, but it is not yet a hard guarantee that European defenders will get timely access when they need it.
Another practical measure is the planned secure testing platform for AI in cybersecurity, to be organised by ENISA and the Joint Research Centre. The platform would allow AI capabilities to be tested in controlled cybersecurity use cases such as vulnerability scanning, remediation, incident response, detection, triage and threat intelligence. The plan also refers to cyber ranges, where AI can be tested against simulated infrastructure without exposing real systems to risk.
This is a sensible approach. Critical infrastructure operators should not be expected to experiment with powerful AI tools directly in production environments. A controlled testing layer can help organisations understand what these tools can actually do, where they fail, and what governance is needed before operational deployment.
The strongest operational part of the plan concerns vulnerability management. The Commission correctly identifies a dangerous imbalance: AI can accelerate vulnerability discovery, but remediation remains slow, fragmented and resource-constrained. This creates a structural advantage for attackers. If attackers can find weaknesses faster than organisations can patch them, traditional vulnerability management cycles will no longer be sufficient.
To address this, the plan calls for Europe’s vulnerability management infrastructure, including the EU Vulnerability Database and CRA-related reporting mechanisms, to become fit for AI-assisted discovery. It also calls for updated coordinated vulnerability disclosure policies and EU-wide risk-based guidance to help manufacturers and users prioritise fixes where the risk is greatest.
The Commission also focuses on open source software, which is deeply embedded in critical infrastructure and enterprise systems. The proposed Critical Open Source Resilience Campaign would map critical open source components and connect projects with sponsors able to provide skilled people, AI tools or other support. This is a positive and realistic move. Many critical open source projects are maintained by small teams, yet their components sit inside public services, industrial systems, healthcare, finance and digital infrastructure.
CyberKreuz analysis shows that the plan also has clear weaknesses. The plan is ambitious, but much of it remains future-oriented. Many actions are scheduled for late 2026 or 2027. The threat, however, is already developing. Boards and CISOs should therefore not treat the action plan as something that will solve the AI-cyber problem for them. It is a policy framework, not an operational shield.
The plan also depends heavily on coordination between the Commission, ENISA, Member States, the JRC, CERT-EU, industry, open source communities and international partners. This is necessary, but complex. Europe already has many cyber instruments: NIS2, DORA, the Cyber Resilience Act, the AI Act, the Cyber Solidarity Act and sectoral initiatives. The challenge is no longer only whether Europe has rules. The challenge is whether those rules can be translated into fast operational capability.
Another concern is access for smaller organisations. The planned testing platform will rely on participants using their own model access keys, tools and resources. Large public authorities, major operators and mature cybersecurity companies may be able to participate effectively. Smaller hospitals, municipalities, SMEs and regional critical infrastructure providers may struggle unless practical support mechanisms are added.
The plan’s investment ambition is also unresolved. The Commission recognises that sovereign frontier AI capabilities may require hundreds of billions of euros and cannot be funded by public money alone. This is an honest assessment. It also shows the scale of the gap. Europe cannot build AI-cyber sovereignty through guidance documents alone. It will require compute, data, talent, procurement, private capital and real demand from critical sectors.
For cybersecurity leaders, the action plan should be read as an early signal of where European cyber policy is moving. AI risk will increasingly become part of supervisory expectations under existing frameworks such as NIS2 and DORA. Patch speed, vulnerability prioritisation, secure AI deployment, supplier dependency and AI skills will move higher on the board agenda.
CyberKreuz assesses the plan as strategically sound but operationally incomplete. Its strongest contribution is that it correctly frames AI-enabled cybersecurity as a sovereignty, resilience and speed issue. Its weakest point is that many of the proposed measures are still voluntary, advisory or dependent on future coordination and investment.
The practical message for organisations: do not wait for EU-level platforms to mature. Review AI-related cyber risk now. Reassess patching cadence. Identify where AI tools are already entering security operations. Test them before sensitive deployment. Include AI misuse scenarios in incident response exercises. Review dependency on non-European AI providers for critical security functions. And ensure that cybersecurity teams are not only aware of AI risks, but trained to use AI safely and effectively.
Europe is preparing its institutional response. Organisations should prepare their operational response now.