In the RapidFort webinar Best Practices for Software Supply Chain Security, Chief Strategy Officer George Manuelian showed runtime profiling results for an NGINX image: 105 of its 152 packages never ran. A scanner reports the CVEs in those 105 packages at the same severity as the CVEs in code that does run. That gap is why container CVE triage has to start with runtime context, before anyone looks at a CVSS score.
Federal policy now points the same way. CISA’s BOD 26-04 sets patch deadlines from four factors: exposure, known exploitation, automatability, and technical impact. CVSS isn’t one of them, and FedRAMP’s matching vulnerability rules take effect December 7, 2026.
This article covers three filters to apply before severity (is the vulnerable code loaded, is it reachable in the way you run the application, and is it exposed), how BOD 26-04 changes patch deadlines, and what evidence a “not affected” decision needs to hold up in an audit.
Why scanner counts overstate container risk
Most scanners work from presence. They inventory what exists in an image, produce a software bill of materials (SBOM), and match each package against known CVEs. They don’t check whether a package is ever loaded into memory or executed.
Across a full container estate, the counts get large. Teams scanning properly for the first time commonly find thousands of vulnerabilities, and tens of thousands isn’t unusual. Some large financial institutions and security companies find hundreds of thousands, and in some cases millions.
Those findings are real vulnerabilities in real packages. In the NGINX example, though, most packages never loaded, so the CVEs in them describe code that sits on disk and never executes in the running application. A queue that treats every finding as equally urgent sends engineers to fix that code while exploitable issues wait.
How to triage container CVEs: three filters before severity
Before a CVSS score sets a finding’s place in the queue, run it through three questions.
Is it loaded?
Runtime profiling records what the application executes and produces a Runtime Bill of Materials (RBOM®) alongside the SBOM. In the NGINX example, the image held 152 packages and 48 CVEs. Profiling showed 47 packages in use and 105 that never loaded.
NGINX can act as a load balancer, a proxy, and more, and most deployments use only part of it. Packages that never load drop to the bottom of the triage queue. They’re also candidates for removal from the image, which companion article on removing unused components in the build pipeline covers.
Is it reachable in the way you run it?
Reachability covers how the application runs as well as network paths. A MongoDB image shown in the webinar carried a medium-severity glibc CVE. The flaw is in code optimized for IBM Power10 processors, and the image was built for AMD64. The CVE is real, and it can’t execute on that architecture.
The same reasoning applies to a kernel CVE reported inside a container image. Containers run on the host’s kernel, so the fix belongs on the host, and the image-level finding doesn’t apply to the container.
Is it exposed?
A vulnerability reachable from the internet is more urgent than one behind several layers of the application. BOD 26-04 builds this into its deadlines: an asset reachable from a public network gets a faster clock, and an asset with unknown exposure counts as public.
Severity becomes useful after these three questions. A CVSS 9.8 in a package that never loads can wait for a scheduled update. A low-severity CVE in loaded, internet-facing code that could give an attacker full control of the container goes to the top of the queue.
How BOD 26-04 replaces severity clocks with four factors
CISA issued Binding Operational Directive (BOD) 26-04, “Prioritizing Security Updates Based on Risk,” on June 10, 2026. It replaces BOD 19-02, which gave federal agencies 15 days to fix critical vulnerabilities and 30 days for high vulnerabilities on internet-accessible systems. It also replaces BOD 22-01, which created the Known Exploited Vulnerabilities (KEV) catalog. BOD 26-04 scores each vulnerability on each asset with four factors:
- Exposure: is the asset reachable from a public network?
- KEV status: is the CVE in the KEV catalog, meaning there is evidence of exploitation in the wild?
- Automatability: can an attacker automate every step of the exploit?
- Technical impact: does exploitation give the attacker partial or total control of the asset?
The 16 combinations map to five deadline tiers: 3 days with forensic triage, 3 days, 14 days, 60 days, and the asset’s next scheduled upgrade. A CVE that is exposed, automatable, and gives total control gets 3 days, whatever its CVSS score. If that CVE is also in the KEV catalog, the agency must also run forensic triage to check whether the asset was already compromised.
The directive binds federal civilian agencies, and CISA has told them to update contracts so their vendors meet the same obligations. On June 16, 2026, FedRAMP published notice NTC-0014, which aligns its Vulnerability Detection and Response (VDR) and Vulnerability Evaluation and Reporting (VER) rules with BOD 26-04. Those rules become mandatory on December 7, 2026, for cloud offerings obtaining or maintaining FedRAMP certification.
Organizations outside the federal market can use the same four factors as a blueprint. Start with the KEV catalog: it’s a better first filter than a severity label.
What a “not affected” decision needs to hold up in an audit
Filtering by runtime context creates a category many teams haven’t managed before, CVEs that are real but don’t affect them. Teams record these as “not affected,” usually in Vulnerability Exploitability eXchange (VEX) format, so scanners stop raising them. CISA’s VEX guidance defines standard justifications for that status. The MongoDB case fits “vulnerable code not present,” and a CVE in an unused NGINX package fits “vulnerable code not in execute path”.
A VEX file that tells auditors to ignore a list of CVEs, with no reasoning attached, doesn’t count as evidence. In FedRAMP and Cybersecurity Maturity Model Certification (CMMC) assessments, teams document each decision in a plan of action and milestones (POA&M) with a justification. Evidence that holds up has three parts:
- A plain-language reason, such as “This CVE affects Power10, and this image runs on AMD64.”
- Supporting references, such as links to the CVE’s National Vulnerability Database (NVD) entry and its CVSS details.
- A verification process behind it. RapidFort builds its advisories by running multiple scanners, examining where their results differ, and having a dedicated team of 50 engineers confirm whether each vulnerability is real. The evidence library ships with the VEX file.
The evidence is only as complete as the profiling behind it. Profiling has to cover every file, package, and process running in the environment, across the OS layer, the framework layer, and compiled binaries. That includes code loaded at runtime through reflection or plugins.
Four questions to ask about your vulnerability backlog
Manuelian closed this part of the webinar with questions any security leader can put to their own program:
- When the scanner reports a critical finding, does anyone check whether it’s meaningful in context?
- How many findings in the current backlog have been triaged or marked “not affected”?
- If you sorted the backlog by reachability instead of CVSS, what would rise to the top?
- Can your team answer that today with tools and a dashboard, or would it be a manual exercise?
If the answer to the last question is “manual,” that’s the gap to close first. CISA has committed to reviewing BOD 26-04 and updating its implementation guidance as necessary, so the rules your triage process has to meet will keep changing.
Prioritize what attackers can exploit
A list of 200,000 vulnerabilities sorted by severity looks rigorous. In practice, engineers end up fixing critical findings in code the application never loads while a reachable low-severity issue waits its turn. Filter by loaded, reachable, and exposed first, back every “not affected” decision with evidence, and your team works from a list of real risks.
Put live clocks on the CVEs that matter. Read how RapidFort Continuous Compliance assigns each container CVE its BOD 26-04 tier and deadline in Turning CISA’s BOD 26-04 Into Live Clocks for Critical Infrastructure.
Latest posts
Eliminate Attack Vectors at the Source
Continuously eliminate up to 99.9% of CVEs without code changes
Products
Use Case
Quicklinks
Stay in touch
Subscribe for product updates and RapidFort newsletter.
Address
440 North Wolfe Road, Sunnyvale, CA 94085
© 2026 RapidFort, Inc.
RapidFort, RAPIDFORT, and RBOM® are registered trademarks of RapidFort, Inc. All other marks and names mentioned herein may be trademarks of their respective companies.





