A standalone system does not get a shorter control list. It gets a different evidence model. When someone tells you the RMF rules do not apply because the box is offline, what they mean is that the automated evidence is gone and they would like the control to go with it. The air gap is not an exemption from NIST SP 800-53. It is one way of implementing boundary protection, and it still has to be described, assessed and defended. What genuinely drops out of a disconnected baseline is narrow: remote access, external connections, nonlocal maintenance and the enhancements built on them. Everything else survives in manual form, which is where standalone authorizations go to die.
Air Gap Is an Implementation, Not an Exemption
NIST SP 800-53 Rev 5 carries 20 control families and one control that decides this argument: PL-11, Baseline Tailoring. PL-11 treats tailoring as a documented decision with a rationale attached to each change, and the actions it recognizes come from SP 800-53B: designating common controls, applying scoping considerations, selecting compensating controls, assigning parameter values and adding controls where risk warrants. Deleting a control because the system is offline is not on that list. CNSSI 4009 and SP 800-37 Rev 2 define tailoring the same way, so claiming a control does not apply is a scoping claim, and a scoping claim has to survive an assessor reading your system description. That is the discipline in the walkthrough of how to build, tailor and defend a control baseline, applied to a harder case.
Confirm you have an information system at all first. A disconnected box inside a weapon system or a test stand may be platform IT, and the PIT versus PIT system question changes the authorization path entirely. Tailoring a baseline you were never supposed to be on is expensive and avoidable.
What Actually Tailors Out of a Disconnected Baseline
There is a published artifact for this, and its status matters before you cite it. The Closed Isolated Network overlay in the NIST Security and Privacy Control Overlay Repository was sponsored by US Army Europe G2, published in October 2020, scoped to low and moderate impact closed isolated networks, and built on SP 800-53 Rev 4. NIST keeps it in the SCOR overlay archive, maintained for archival purposes only, so it is not current authority. It is still the best public record of the tailoring an authorizing official has already accepted for an enclave that, in the overlay’s own words, does not connect to any other network.
The arguments that hold
The tailored-out set in that overlay is tight, and every item in it is connection dependent. AC-17 Remote Access and its enhancements go. AC-19 and AC-20 go with mobile and external devices, AC-21 and AC-22 with information sharing and publicly accessible content. CA-3 Information Exchange goes, because there is nothing to exchange with. MA-4 Nonlocal Maintenance goes, which is the claim most often simply true: nobody is dialing in. AU-7, AU-10, the IA-8 controls for non-organizational users, the SA-9 external services controls and the bulk of the SC-7 enhancements also drop, while SC-7 itself survives with only its external-accessibility items marked not applicable. The pattern: a control tailors out when the asset it protects does not exist inside the boundary, not when the evidence got inconvenient.
The arguments that do not
Here is the set I watch get claimed not applicable, then reversed:
- RA-5 Vulnerability Monitoring and Scanning. The overlay keeps it, sets a scan at least every 120 calendar days plus one when new vulnerabilities are reported, and blesses a standalone laptop scanner with security office approval.
- SI-2 Flaw Remediation. Kept, with security-relevant updates installed within 90 days of release for a single-site enclave and 60 days for a multi-site one, after testing and approval under the configuration management plan.
- AU-6 Audit Record Review. Kept, on a 90-day cycle for single-site and 60 days for multi-site. Logs do not stop because nothing can reach the box.
- CM-6 Configuration Settings. Kept and strengthened: enclave-specific baselines in the most restrictive mode consistent with operations, with CM-7(5) deny-all, permit-by-exception software control.
- IA-2 and IA-5. Local accounts are still accounts, and a shared administrator login is the same finding offline as anywhere else.
- The whole MP family. An air gap concentrates risk into removable media, it does not remove it.
- CP-9 and CP-10. Backup and recovery get harder offline, not optional.
Writing not applicable down forty rows of the control list because the system is air gapped is the fastest way I know to turn one tailoring conversation into forty findings.
Scanning and Patching With No Network
Offline configuration assessment
SCAP Compliance Checker, developed and released by NIWC Atlantic, runs fully disconnected. No cloud console, nothing to phone home to. You feed it SCAP content, which bundles XCCDF, OVAL and CPE, and it produces XCCDF and ARF results plus a checklist you open in STIG Viewer for the rules automated content cannot evaluate. It will not fetch the benchmark for you. Content moves quarterly, so your sustainment plan needs a written process for carrying the current benchmark in and recording which version produced which result.
Automated content covers only the checkable subset of a STIG. On a connected system the manual rules are a rounding error. On a standalone system they are the assessment, and the artifact is the reviewed checklist with a reviewer, a date and finding detail for every manual rule.
Offline vulnerability scanning
ACAS inside an air gap is a full deployment, not a reach-in. Tenable’s guidance for air-gapped environments is to stand up a Security Center instance with its own scanners inside each isolated network, and to use offline repositories if that data has to reach another instance. Plugin and feed updates follow four steps: download the files on a connected system, verify them against the published checksums, move them in, upload them through the browser or the API. Tenable recommends updating plugins and feeds at least monthly, upgrading the platform at least yearly, and running comprehensive scans after each plugin update. A handful of plugins need internet access and will never return a result offline, the botnet membership check being the standard example. The credentialed scanning mechanics are the ones in the walkthrough of ACAS and Nessus for ISSOs.
Record the plugin feed date behind each scan, and write the offline update procedure into the system security plan as the RA-5 implementation statement. “We scan quarterly” says nothing when the hard part is the courier.
How the patch media gets in
The patch path is the control. Describe it end to end: who retrieves the update from the authoritative source, what verifies the hash, what media it is written to, who approves the change at the configuration control board, what test system it lands on first, who carries it, who logs the entry and the exit, and what happens to that media afterward. Every one of those steps is already a control. The transfer procedure is not overhead wrapped around your controls. Those steps are CM-3, MP-5, MP-7 and SI-2 under different names.
Media Handling Is the Real Risk Surface
On a connected system the interesting attack path is the network. On a standalone one it is whatever is in somebody’s pocket. The Rev 5 Media Protection family is short and all of it applies: MP-1 Policy and Procedures, MP-2 Media Access, MP-3 Media Marking, MP-4 Media Storage, MP-5 Media Transport, MP-6 Media Sanitization, MP-7 Media Use and MP-8 Media Downgrading. The overlay strengthens several: need-to-know restriction on media access, a marking exemption that holds only inside a controlled area, secure storage, transport under CNSSP No. 26, and media use restricted to what the enclave design documentation allows. MP-6 is where standalone programs lose points, because the disposal evidence is a paper trail and paper trails go missing. Method, operator, verification and certificate are all part of the artifact, and the walkthrough of wipe, degauss, destroy and the certificate you have to keep is the version to hand your property book officer.
Your transfer and disposal procedure has to produce evidence:
- A signed, dated media control log: control number, classification, who drew and returned it.
- An approved list of authorized media types and what is prohibited, mapped to MP-7.
- Hash verification records for content brought in, tied to its authoritative source.
- Configuration control board minutes approving the change the media delivered.
- A sanitization or destruction record for every piece of media leaving the controlled area, with the certificate retained per your records schedule.
- A current CM-8 component inventory covering the media itself where policy treats it as a component.
Physical and Personnel Controls Carry More Weight
Remove the logical boundary controls and the boundary becomes a door and a cleared person. The overlay reflects that by tightening the physical and maintenance families. PE-3(1) pushes access enforcement down to the spaces holding system components, not the building perimeter. PE-6 sets access log review at 90 days for a single-site enclave and 60 for a multi-site one. PE-16 covers delivery and removal, the same chokepoint as your media path. MA-2 requires ISSM and ISSO approval before equipment leaves the site, and MA-3(3) exists to stop maintenance equipment walking out with data on it.
Personnel security tightens the same way, with parameters specific enough to fail an inspection on: access removal within five working days for a voluntary termination and the same day for an involuntary one, transfers and third-party changes actioned within one working day, sanctions reported to the enclave security administrator within one working day. The mechanics are laid out in the piece on PS controls and the termination nobody reported. On a standalone system there is no automated deprovisioning job to blame. There is you, a list and a date.
The Not-Applicable Trap
The single most expensive move in standalone RMF is marking a control not applicable when it has a perfectly good manual implementation. You convert something you already do into a finding, because the record now says you do not do it. PL-11 wants a rationale for each change to the baseline, and “we perform this manually, on this cadence, and here is the log” is a strong one. “Not applicable, no network” against AU-6 is an invitation.
The test before you write not applicable: name the asset the control protects, then say why it does not exist inside your boundary. If you can name the asset, the control applies and you owe an implementation statement. If the honest answer is that your implementation is manual, partial or slower than the parameter, that is a POA&M item, and one with a milestone beats a not-applicable the assessor reverses in week two.
Before your next assessment, do three things. Walk the boundary and confirm the system description matches the room, media and maintenance laptops included. Write the one-sentence asset test for each control you marked not applicable. Then pull the last four quarters of manual STIG reviews, scans, audit reviews and access log reviews, and confirm each carries a reviewer and a date. None of this is harder than connected RMF, just less forgiving, because nothing fills itself in. For the control-by-control version with the artifact names and the build order, that is what the RMF Checklist is for. It assumes you are the one carrying the evidence, which on a standalone system you are.


Leave a Reply