CMMC and FedRAMP Diagrams: Boundary vs. Network vs. Data Flow

Somebody has told you that you need three diagrams: an authorization boundary diagram, a network diagram, and a data flow diagram. You have drawn one picture that tries to be all three, and it has come back with comments.

This is one of the most common places a compliance effort stalls, and the reason is not that the drawing was bad. The three diagrams answer three different questions, and a picture that answers all of them at once usually answers none of them clearly.

It is also worth knowing — before you spend a week in Visio — which of these a rule actually requires, as opposed to a template, a playbook, or an assessor's preference. The answer is narrower than the industry's confidence suggests, and in late 2026 it is moving in two directions at once.

The three questions

Strip away the vocabulary and each diagram exists to settle one argument.

The authorization boundary diagram answers: what are you accountable for? It is less a technical drawing than a statement of responsibility. Everything inside the line is something you are claiming to secure and will be assessed on. Everything outside it is someone else's problem, and you will be asked to justify why.

The network diagram answers: how is it connected? This is topology — segments, firewalls, where traffic can and cannot go. It is the engineering view, and it is what someone reads when they want to know whether the boundary you drew is enforced by something other than good intentions.

The data flow diagram answers: where does the regulated data actually go? Not where your systems sit, but the journey a specific kind of data takes — who touches it, where it rests, where it is encrypted, where it leaves your control. A network diagram shows the roads. A data flow diagram shows the trip.

The conflation that causes the most rework is between the first two. People draw a network diagram, put a red box around part of it, and call it a boundary diagram. That is a reasonable starting point, and the boundary is often drawn as an overlay. But the boundary has to account for things a network diagram does not naturally show: external services you depend on, infrastructure you do not own, and the points where your responsibility hands off to somebody else's.

What each framework actually requires

CMMC requires a network diagram, and only a network diagram. This is binding regulatory text rather than guidance. At Level 2, 32 CFR 170.19 directs that four of the five asset categories — CUI Assets, Security Protection Assets, Contractor Risk Managed Assets, and Specialized Assets — each be documented "in the network diagram of the CMMC Assessment Scope." Out-of-Scope Assets carry no documentation requirement at all, though you should be ready to justify why they cannot process, store, or transmit CUI. Level 3 works the same way across its four categories. Level 1 has no documentation requirement whatsoever — no diagram, and a system security plan is recommended rather than required.

CMMC does not require a data flow diagram. The phrase does not appear in the Level 2 Scoping Guide, and it does not appear in the regulation. It surfaces in the Level 2 Assessment Guide only inside a list of document types that might serve as evidence — a list the guide itself describes as "not exhaustive or prescriptive," adding that an organization may not have these specific documents and that others may be reviewed instead. Draw one anyway, for reasons we will come to. Just know you are meeting an expectation, not a requirement.

NIST SP 800-171 requires no diagram. The system security plan requirement asks you to "develop, document, and periodically update system security plans that describe system boundaries, system environments of operation, how security requirements are implemented, and the relationships with or connections to other systems." All text. No picture. This is precisely why CMMC had to add the network diagram in its own regulation — the underlying standard does not contain one.

Worth knowing for the inevitable argument: Revision 2 of 800-171 was formally withdrawn in May 2024 when Revision 3 superseded it, yet Revision 2 remains the CMMC baseline, because 32 CFR 170.14 incorporates it by reference and states that the Level 2 requirements are identical to it. A withdrawn publication is still the one you are assessed against.

The RMF itself requires no diagram. NIST SP 800-37 treats the authorization boundary as a set of system elements and never mandates depicting it graphically.

The thing that changed in June, which almost nobody has noticed

For twenty years, the companion guide on how to actually write a system security plan — SP 800-18 Revision 1, published in 2006 — asked for the system environment in "one to three paragraphs," plus a list of primary hardware, software, and communications equipment. No diagram anywhere. The entire diagram culture around ATOs was convention built on top of a standard that asked for prose.

That ended on 30 June 2026, when Revision 2 superseded it. Revision 2 names diagrams as plan elements outright: Environment of Operation Diagrams, with the instruction to "include system diagrams that clearly depict the components of the system architecture within the authorization boundary," and beneath that, System/Subsystem Diagrams, System/Network Architecture Diagrams, and Data Flow Diagrams. The phrase "one to three paragraphs" is gone.

So the convention is now written down. If you have been drawing these for years because an assessor expected them, the guidance has caught up with you.

And FedRAMP is going the other way

FedRAMP under Rev 5 is the most prescriptive diagram regime in federal compliance, and it is where the three-diagram convention comes from. All three live in Section 8 of the system security plan, "Illustrated Architecture and Narratives." The authorization playbook states that to understand the level of detail that "must be provided in the ABD, DFD and network diagram," you review the instructional text there. The boundary is then validated against the inventory during the independent assessment.

But FedRAMP has said it will retire that requirement. A notice published in March 2026 states plainly that "FedRAMP will not require diagrams or illustrations after the transition to the Minimum Assessment Scope," and that there is "no expectation in the Minimum Assessment Scope of a traditional Authorization Boundary Diagram that contains every single service and flow in a single diagram." Providers get flexibility to present their structure across multiple levels instead.

What replaces it is an obligation to "clearly identify, document, and explain information flows and security categories" for all in-scope resources — with no format specified. A diagram satisfies that. So does structured prose. So does machine-readable data, which is plainly the direction of travel: the Consolidated Rules for 2026 drop the system security plan in favor of a Security Decision Record, required in both human-readable and JSON form.

Two practical consequences. If you are maintaining a legacy Rev 5 package, the three-diagram expectation still governs what is already in review — though FedRAMP now publishes that playbook as reference-only legacy material. If you are starting fresh, you are building toward a regime where the diagram is a communication tool rather than a deliverable. Either way, do not buy a diagram package from anyone claiming FedRAMP will demand one indefinitely.

How they are the same

Underneath, all three diagrams do one job: draw a line between what is in and what is out, along three different axes — accountability, connectivity, and data movement. They should agree with each other. Where they do not, you have usually found a real problem rather than a drafting error.

And they are judged by the same two tests, whichever framework you are under.

Does it match reality? Nobody evaluates the drawing; they evaluate whether the drawing is true. The probing happens at the edges: how is an out-of-scope workstation actually prevented from reaching the server holding CUI? If the answer is a VLAN with no enforcement behind it, the segmentation is decorative and the other side comes into scope. One jump box bridging two environments can pull an entire network back in.

Does it agree with everything else you submitted? This is where most packages come apart. The diagram has to match the asset inventory, the plan narrative, and the control descriptions. FedRAMP is explicit that the boundary is validated against the inventory. CMMC is not — the guides require both an inventory and a network diagram but never say the two must reconcile, which is exactly why assessors check it. A working heuristic from practitioners: if it is mentioned in the system security plan, it should be on the diagram.

What gets a diagram sent back

FedRAMP has published draft Review Initiation Checks — optional self-assessment checklists meant to catch what most often needs rework before a package goes in. They are not the standard a package is formally measured against, but they are the most concrete public list available and a fair proxy for what any careful reviewer looks for.

  • It is not legible. One check asks whether the resolution is "fully legible, including all characters being clearly distinguishable," and whether text in the diagrams is searchable. If a reviewer has to zoom, you have already cost yourself time.
  • Flows are drawn to groups rather than components. "In general, depicting flows to groups such as VPCs is insufficient."
  • External services are not marked. Each should be identified as either authorized or lacking authorization.
  • Encryption is invisible. Data flow diagrams are expected to show the encryption status of data in transit and at rest, and to include the components where encryption can terminate — load balancers and firewalls among them.
  • Backup and alternate-site flows are missing. The boundary diagram should depict the alternate processing and storage site, and the data flow diagrams should show those flows. Note that FedRAMP distinguishes backup from mirroring or replication.
  • Ports and protocols do not reconcile. This check is not against the diagram at all — it compares the plan's ports and protocols table against what the vulnerability scans actually found open. A mismatch there undermines every drawing in the package.

One more, from FedRAMP's older readiness guidance rather than the current checks: a legend. The Rev 4 readiness assessment guide asked for a high-resolution boundary diagram with one, and reviewers still ask. It costs nothing to include.

The pain points nobody resolves for you

Security protection assets get left out, then pulled back in. The CMMC scoping guide defines these as assets providing security functions or capabilities within your assessment scope, and names cloud-based security solutions, hosted VPN services, SIEM solutions, security operations centers, and the people who run them. Assessors we have heard from push the definition further: if a tool helps you meet a security requirement, it is probably a security protection asset even if it never touches CUI. The one organizations most often forget is the ticketing system, because it accumulates incident records and vulnerability data. Leaving these off does not shrink your scope. It delays your assessment while somebody puts them back.

The enclave that does not match how people work. A pattern assessors describe: an organization buys a tightly scoped enclave, then handles CUI outside it because the enclave cannot do something the business genuinely needs. Printing is the example we hear most often. The result is a diagram that accurately describes a system nobody fully uses. The fix sits upstream of the drawing — scope to the work people actually do.

The leftovers category. Contractor Risk Managed Assets are for assets that could technically touch CUI but are managed not to. In practice the category gets used as a container for whatever did not fit elsewhere, and assessors we have spoken to say they recognize the pattern. A CRMA still has to appear in the inventory, in the plan, and on the network diagram, positioned relative to the boundary, with a documented rationale. And it does not travel upward: the category does not exist at Level 1, and at Level 3 it is not relaxed but removed — CRMAs are treated as CUI Assets and assessed against all Level 3 requirements. A scope built on CRMA does not simplify when you move up. It converts.

The cloud question, which is genuinely unsettled. Whether a given cloud service sits inside or outside your boundary, and what proof you need from the provider, is still contested among competent practitioners. Two things hold. Services provided by an external provider are in your assessment scope, and you are still assessed on your own infrastructure connecting to them — you determine which requirements are implemented and which are inherited, and where an enterprise tool would invalidate your implementation, that requirement cannot be inherited at all. And an authorization does not necessarily cover every feature of a product; add-on modules can sit outside it. A customer responsibility matrix entry saying the work is handled by a managed service provider, with no further detail, is a finding waiting to happen.

Drift, which beats everyone eventually. The hard part is not drawing the diagram, it is keeping it true. Emergency firewall rules, a temporary VPN, infrastructure changed by a pipeline that never touched the documentation, the fix at 2am nobody wrote up. When a reviewer says a diagram must be current, they mean it reflects the network today — not that it was drawn recently.

What to do about it

Draw the data flow first, even where no rule requires it. You cannot sensibly decide what belongs inside a boundary until you know where the regulated data goes, and the exercise reliably surfaces a hop nobody remembered — a reporting tool, a backup target, a personal device someone uses to review documents at home.

Then draw the network diagram. Then overlay the boundary. One diagram per question, cross-referenced, each readable on its own.

Keep the inventory and the diagram in lockstep, because that comparison is cheap for an assessor to make and they make it early. And give someone standing ownership of updating all of it when the environment changes, because the alternative is finding the gap during an assessment, when it stops being a documentation problem and becomes a finding.

If a CMMC or federal requirement has landed on your desk and you are not sure which of this applies to you, book a 30-minute deal-readiness call. If it arrived inside a customer security questionnaire, Security Questionnaire Rescue starts at $1,500. For how the current CMMC pause affects any of this, see where CMMC Level 2 stands right now.

Written October 2026, citing 32 CFR 170.14 and 170.19; the CMMC Level 1, 2 and 3 Scoping Guides and the Level 2 Assessment Guide (v2.13, September 2024); NIST SP 800-171 Rev 2; NIST SP 800-37 Rev 2; NIST SP 800-18 Rev 2 (June 2026); the FedRAMP CSP Authorization Playbook (v4.2, November 2025); FedRAMP notice NTC-0009 (March 2026); and FedRAMP's draft Review Initiation Checks. Requirements in this area are moving quickly — confirm the current position for your own situation before acting on it.

Is security holding up a deal?

Book a deal-readiness call