23/07/2026
The European Commission publishes its Action Plan on Cybersecurity and Artificial Intelligence
The publication of the European Commission’s Action Plan on Cybersecurity and Artificial Intelligence comes at a time when the conversation on AI is beginning to move away from the comfort of regulatory debate and into the realm of real-world security. Europe has built an ambitious regulatory framework: the AI Regulation, NIS2, DORA, the Cyber Resilience Act and the Cyber Solidarity Act. The assessment, however, is clear: having good rules does not guarantee preparedness if critical capabilities are developed elsewhere, if attackers are advancing faster, and if organisations continue to manage cybersecurity using processes designed for a bygone era.
Artificial intelligence and cybersecurity
Frontier AI changes the scale of the problem. It enables threats to be detected, vulnerabilities analysed, risks prioritised and responses delivered at a speed impossible for human teams working alone. But those same capabilities can also automate attacks, make criminal operations cheaper, improve phishing campaigns, accelerate the exploitation of vulnerabilities and lower the technical barrier to entry for less sophisticated actors. The question is no longer whether AI will have an impact on cybersecurity. The question is who will use it first and with what controls.
The Commission’s plan is right to frame the problem on three levels. The first is the assessment of advanced models before they reach the market or are deployed in sensitive environments. It is not enough to rely on suppliers’ declarations. If a model can help exploit systems, uncover vulnerabilities or generate offensive capabilities, Europe needs its own capacity to assess that risk and understand what it is putting into circulation. This is a discussion not only of compliance, but also of technical sovereignty. An authority that cannot assess what it is supervising ends up depending on the technology’s manufacturer.
The second issue is access. Many advanced capabilities are in the hands of non-European suppliers and subject to decisions that lack transparency: who has access, with what restrictions, under what criteria and for how long. It may be reasonable to restrict access for security reasons, but it is not acceptable for critical operators, public authorities or European cybersecurity providers to depend on opaque processes to protect essential infrastructure. The European Blueprint announced in conjunction with ENISA could prove significant if it regulates such access without creating excessive bureaucracy.
The third aspect is operational: test before deployment. The idea of secure platforms and cyber ranges for testing AI in real-world cybersecurity scenarios is sound, because not every model that works in a laboratory setting will function within a critical organisation. We need to understand how it behaves in the face of false positives, complex incidents, legacy systems, regulatory restrictions and sensitive data. AI applied to cybersecurity cannot go into production merely as a commercial promise.
Discovering vulnerabilities
But the document’s most important message lies in something more fundamental: speed. The Commission emphasises that AI is changing the economics of vulnerability discovery. Attackers will be able to find more flaws, sooner and at a lower cost. If remediation, patching and internal coordination remain the bottlenecks, the imbalance will widen. Many organisations lose the battle not because of a lack of tools, but because of slow processes, incomplete inventories, unmanaged dependencies and delayed decisions.
That is why the plan links AI with NIS2, DORA and the Cyber Resilience Act. It is not a question of adding another layer of regulation, but of ensuring the fundamentals work: risk management, supplier governance, security by design, vulnerability management, business continuity, incident response and accountability of senior management. In the age of AI, these fundamentals cease to be merely a compliance checklist and become a competitive advantage. Any organisation that cannot patch quickly, prioritise effectively and understand its actual exposure will be more vulnerable, even if it has formally sound policies.
It is also right for the Commission to look at open-source software. A significant part of the digital economy relies on components maintained by small communities, which are sometimes under-resourced but present in critical infrastructure. If AI accelerates the discovery of vulnerabilities in these components, Europe must also help to accelerate their maintenance, funding and remediation.
The big question is whether Europe will be able to act at the pace demanded by the risk. The document discusses assessment, access, testing, training, investment, AI Factories, sovereign capabilities and international cooperation. But success will depend on organisations understanding that AI in cybersecurity is not just another tool, but a paradigm shift. It necessitates a review of defence architecture, relationships with suppliers, vulnerability management, the role of the board and the preparedness of teams.
Europe has demonstrated its ability to regulate. It must now demonstrate its ability to implement. In cybersecurity, being late rarely comes without a cost. And in the age of AI, the difference between being late and not arriving at all can be measured in days, not years.
