Your supplier's patch cadence is your patch cadence – making it measurable under NIS2
If you have outsourced operations, hosting or application management, your provider's patch cadence is your patch cadence – whatever your own security policy says. Most organisations know this. Far fewer can say how many days it actually took last quarter for a critical fix to reach production in a system someone else runs for them.
NIS2 already puts this on your desk
NIS2 does not let you outsource the responsibility along with the work. Article 21 lists the risk-management measures essential and important entities must take, and two of them meet exactly here:
- Supply chain security – "including security-related aspects concerning the relationships between each entity and its direct suppliers or service providers" (Article 21(2)(d)).
- Security in acquisition, development and maintenance – "including vulnerability handling and disclosure" (Article 21(2)(e)).
Article 21(3) adds that entities must take into account the vulnerabilities specific to each direct supplier and the overall quality of their suppliers' cybersecurity practices. In other words: "the supplier applies security updates in accordance with best practice" is not something you can file and forget. You are expected to know what it means in practice.
Contract terms that can actually be measured
The problem is rarely bad intent. It is that patching is described in words nobody measures. If we were reviewing an operations contract today, we would ask for:
- Time to patch, by severity and exploitation status. Separate targets for "actively exploited" and "critical but not exploited", measured from the day a fix is available to the day it runs in production – not to the day it was scheduled.
- A named source of truth for "actively exploited". For example ENISA's EU Vulnerability Database, which has a dedicated view of exploited vulnerabilities, so there is no debate about which list counts.
- Evidence, not attestation. A regular report of open critical findings per system, with their age, produced from the provider's scanner rather than written by hand.
- A defined exception process. Some systems cannot be patched quickly. That is acceptable when it is recorded, has compensating controls and an expiry date – not when it is silent.
- A right to verify. Read access to scan results or an agreed independent check, and a clause that links repeated misses to the exit plan.
None of these is exotic. Together they turn "best practice" into a number you can follow up in a governance meeting.
What a platform should do so patching is not a project
Contracts set expectations; platforms decide whether those expectations are realistic. On container platforms, slow patching is usually a symptom of a delivery path where every update is a manual change. The pattern that works:
- Rebuild, don't patch in place. Base images and dependencies are updated by automation (for example Renovate or Dependabot) as pull requests, and images are rebuilt and rescanned on a schedule, not only when the application changes.
- Know what runs where. An SBOM per image and scanning in the registry mean that when a new vulnerability is published, finding the affected workloads is a query, not an investigation.
- Let the cluster enforce the floor. An admission policy (Kyverno or OPA Gatekeeper) can refuse images that are unsigned, come from an unapproved registry or fail the agreed vulnerability threshold, so an old image cannot quietly return.
- Treat nodes and the cluster version as patchable too. Node images rotated on a schedule, and Kubernetes minor upgrades planned within the upstream support window, so the platform itself is never the reason a fix waits.
- Change through Git. When updates arrive as reviewed pull requests applied by a pipeline or GitOps controller, the time from fix to production becomes visible – exactly the number the contract asks for.
The trade-off is honest: this takes engineering effort up front, and it requires tests good enough that teams trust an automated update. Without those tests, automation just moves the risk from "unpatched" to "broken".
The signal is getting better – the deployment is still yours
Since 11 September 2026, manufacturers of products with digital elements must report actively exploited vulnerabilities and severe incidents through ENISA's Single Reporting Platform under the Cyber Resilience Act. That should improve what we know about exploitation. It does not change how fast your provider deploys the fix. That part is still contracts, platforms and follow-up.
Where to start
Ask your operations provider one question: for each critical system, how many days passed between the last critical fix becoming available and it running in production? If the answer takes more than a day to produce, that is your first finding.
Sources
- EUR-Lex: Directive (EU) 2022/2555 (NIS2), Article 21
- ENISA: The CRA Single Reporting Platform is launched (11 September 2026)
- ENISA: EU Vulnerability Database (EUVD)
- The Register: Report on the EUVD becoming fully operational (13 May 2025)