A joint statement from the NCSC, AIVD, MIVD and other Dutch public-sector organisations gives a grounded account of how AI is changing cyber risk. Attackers can use accessible AI models to automate parts of their work, find vulnerabilities more quickly and move from discovery towards exploitation with less manual effort. The agencies warn that weaknesses already present in organisations will become harder to leave unresolved as this capability spreads.
Marc Kuipers, the Netherlands’ National Coordinator for Counterterrorism and Security, captures some of the urgency behind the statement: “We do not want to wait until we see the consequences of this at scale in cyber incidents” (translated from Dutch). Organisations are being asked to prepare before those effects become widespread, rather than after attacker behaviour has already changed around them.
Matthijs van Amelsfort, Director General of the NCSC, is more specific about where that pressure comes from. According to him, “AI automates the attack chain, from finding vulnerabilities to exploiting them” (translated from Dutch).
Poor asset visibility, delayed patching, unnecessary internet exposure and incomplete monitoring already give attackers opportunities. AI can reduce the amount of effort required to search for those opportunities and act on them. The NCSC consequently puts familiar measures at the centre of its advice: effective patch management, a smaller attack surface, logging and monitoring, and a clear view of the IT environment.
The time available to defenders is getting shorter
Patch management has always involved a race between disclosure, remediation and exploitation. AI can accelerate several parts of the attacker’s work, including reconnaissance, vulnerability analysis and the development or adaptation of attack techniques.
The NCSC describes the effect in unusually direct terms, warning that weeks can become days and days can become hours when organisations are deciding how quickly to update.
Those timelines are difficult to reconcile with security processes that only take a fresh look at the environment periodically. An annual penetration test can provide detailed assurance about a defined environment at a particular point in time, while scheduled vulnerability assessments provide regular visibility into known weaknesses. Between those assessments, new applications are deployed, cloud services are configured, domains are registered, credentials are exposed and vulnerabilities are disclosed.
Internet-facing infrastructure can also emerge outside established processes, particularly as companies grow, make acquisitions or allow individual teams to deploy services independently. Every delay between a change taking place and the security team seeing it creates additional time in which an attacker may get there first.
Hadrian’s own research gives some indication of the volume behind that problem. Our recent analysis of vulnerability disclosures found an average of 195 new CVEs published every day, putting 2026 on course for more than 71,000 disclosures if that rate persisted. July subsequently produced 9,771 CVEs in a single month.
Security teams therefore have to respond to faster attacker activity while the amount of information demanding their attention continues to grow.
More vulnerabilities make prioritisation harder
More disclosure does not translate directly into more exploitable exposure for every organisation. A vulnerability may affect software the company does not use, a system that is not externally reachable or a configuration in which exploitation is impractical.
Hadrian’s 2026 research across more than 300 organisations illustrates how much the picture can change once those factors are tested. Only 0.47% of risks detected by vulnerability scanners were ultimately verified as exploitable and requiring action in the environments analysed. That figure reflects Hadrian’s dataset rather than a universal scanner false-positive rate, but it shows the scale of the filtering problem security teams face.
Teams need enough context to establish whether affected technology is present, whether an attacker can reach it and whether the weakness leads somewhere meaningful. As the NCSC’s timeline compresses from weeks towards days and hours, producing that context quickly becomes part of the response problem.
Known exploitation data helps, although it can arrive well after a vulnerability becomes relevant. On average, there are a median 57 days between disclosure and later inclusion in CISA’s Known Exploited Vulnerabilities catalogue. Separate research cited in our benchmark found evidence of exploitation on or before CVE publication for 32.1% of vulnerabilities later included in KEV during 2025.
Waiting for a strong external signal can therefore consume much of the time available to respond. At the same time, hundreds of new disclosures each day make treating every new CVE with equal urgency impossible. Current asset information, exposure context and evidence about practical exploitability become increasingly valuable when teams are deciding where limited remediation capacity should go.
Security programmes need to see changes sooner
The NCSC asks boards to reduce their attack surface, replace outdated systems, maintain effective patch management and prepare for compromise. Delivering on those recommendations depends heavily on how quickly organisations can see changes in their own environments and understand what those changes mean.
An internet-facing service that appears outside the normal inventory needs to be discovered while the information is still useful. Newly disclosed vulnerabilities need to be connected to the systems actually exposed. Findings need enough context for teams to decide what deserves immediate attention, and remediation needs to be checked afterwards.
Long intervals between these activities create periods where organisations are working from an incomplete picture. As AI allows attackers to search, analyse and act across more targets with less manual effort, those periods become easier to exploit.
The NCSC’s warning therefore lands most heavily on organisations that already struggle with the basics. AI gives attackers greater capacity to take advantage of old vulnerabilities, forgotten assets and slow security processes, while giving defenders less time to discover and resolve them. Strong security fundamentals remain useful precisely because they reduce the number of opportunities that faster attackers have available.
If you want to see what that acceleration looks like in public vulnerability data, Hadrian’s Exposure Clock lets you explore how vulnerability disclosure and known exploitation have changed over any period you choose, including the time since your organisation’s last assessment.
