Skip to content

DRAFT Policy - Secure SDLC Policy - #50

Open
knikolla wants to merge 3 commits into
CCI-MOC:mainfrom
knikolla:policy/sdlc
Open

knikolla wants to merge 3 commits into
CCI-MOC:mainfrom
knikolla:policy/sdlc

Conversation

@knikolla

@knikolla knikolla commented Oct 29, 2026 •

Copy link
Copy Markdown

Policy defining secure software development lifecycle practices for MOC.

Publishing an early draft with several open questions still in there for early feedback.

@github-actions github-actions Bot added the ops-policy This is a policy document label Oct 29, 2026
@knikolla knikolla changed the title Secure SDLC Policy DRAFT Policy - Secure SDLC Policy Oct 29, 2026
Comment thread policies/2026-10-27-sdlc/README.md Resolved
Comment thread policies/2026-10-27-sdlc/README.md Outdated
Comment on lines +3 to +9
**Document Owner:** [TBD]
**Approved By:** [TBD]
**Review Cycle:** [TBD]
**Effective Date:** December 1, 2026
**Document Version:** DRAFT
**Parent Document:** System Security Plan (SSP): [SSP link: TBD]
**Target Compliance Framework:** NIST SP 800-171 Rev 2 (Protecting CUI in Nonfederal Systems)

@larsks larsks Oct 29, 2026 •

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If these are attributes we want for every policy, we should add them to the YAML frontmatter section (and update the configuration to render them in a useful fashion). We shouldn't be making up per-document metadata tables.

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Valid question. I'm not sure what all of this is required. This was initially a Google Doc that predated the restructuring of this repo. I'd rather keep as little of it is necessary as possible and I hope the review process of this policy will answer what is required and whether we should update the frontmatter template.

@knikolla

Copy link
Copy Markdown
Author

@larsks i'll apply the correct front matter, thanks!

Comment thread policies/2026-10-27-sdlc/README.md Resolved
Comment thread policies/2026-10-27-sdlc/README.md Resolved
Policy defining secure software development lifecycle practices
for MOC.
@bu-renm

bu-renm commented Oct 29, 2026

Copy link
Copy Markdown
Contributor

Recording some custom software components this SDLC policy applies to:

  • COSI driver
  • Invoicing
  • Coldfront
  • Ansible-switches custom filter code

Comment thread policies/2026-10-27-sdlc/README.md Resolved
Comment thread policies/2026-10-27-sdlc/README.md Outdated
### 1.2 CUI Boundary, Deployment Model & External Systems

* **Platform:** The platform consists of an on-premises OpenShift cluster and foundational services (for example Keycloak and LDAP) on Amazon EKS in AWS. Both are within the CUI boundary. The SSP describes the architecture, the boundary, and the network path between them.
* **Deployment Model:** CI/CD does not deploy to production. Jobs running inside the CUI boundary, pull approved configuration from the public and private GitHub repositories and apply it to production, including the on-premises OpenShift cluster (see Section 6.1).

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ArgoCD is our continuous deployment (CD) tool for our OpenShift clusters, and it does deploy to production.

In other cases -- moc-aws, moc-github, etc -- our GitHub CI/CD workflows do deploy to production.

We could potentially alleviate some of this by setting up on-premise github runners, which probably requires an enterprise license. Or we could just stand up our own on premise CI/CD tooling separate from github (e.g., Jenkins or similar), rather than relying on github's ci workflows (which would be a shame from a usability perspective).

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

For purposes of compliance I don't think we'll be able to keep the GitHub CI/CD workflows running on GitHub infrastructure deploying into prod, as that would bring the whole of GitHub into the CUI boundary.

If we can use on-premise runners once we get access to GitHub Enterprise and just change where things run that would be great, rather than having to maintain a separate system.

I'll update the deployment model to mention ArgoCD for deploying Kubernetes manifests into OpenShift. Does this apply to the EKS cluster too?

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Does this apply to the EKS cluster too?

We are using argocd to manage the moc-services cluster. Is that what you were asking?

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, I'll update the text.

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

None of these changes are planned for MOC 1.5. If changes need to happen after that, IS&T will be leading the effort after the Enterprise license is secured.

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Updated the text to reference to ArgoCD and on-prem GitHub runners, rather than "Jobs"

Comment thread policies/2026-10-27-sdlc/README.md Resolved
Comment thread policies/2026-10-27-sdlc/README.md Resolved
Comment thread policies/2026-10-27-sdlc/README.md Outdated
* **Repositories:** Source code and configuration repositories are hosted under GitHub Enterprise. Configuration is split into public and private repositories. Public repositories store public manifests with templated placeholders for sensitive information such as IP addresses. Private repositories store the sensitive configuration. Public repositories must not contain CUI, credentials, or environment-specific sensitive configuration (for example account identifiers, IP addresses, network topology, or realm and directory configuration). Private repositories must not contain CUI or credentials. [TBD: state who can read the private repositories, and how public repositories are checked for placeholders only (for example CI scanning or a pre-publication review).]
* **Authentication:** Multi-Factor Authentication (MFA) is enforced across the entire GitHub organization for all members.
* **Authorization:** Role-Based Access Control (RBAC) is managed via GitHub Teams, granting least-privilege access to specific project repositories.
* **Actions on Forks:** GitHub Actions workflows never run on forks when they contain secrets. Where workflows do contain secrets, those secrets are never for anything within the CUI boundary.

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Maybe it's worth noting that this isn't just our policy; it's enforced by GitHub (specifically: workflows running against pull requests from forked repositories cannot access repository secrets).

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'll add that as a comment. I was not assuming familiarity form a possible assessor/auditor into how GitHub works.

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Reworded to say GitHub blocks this.

Comment thread policies/2026-10-27-sdlc/README.md Resolved
Comment thread policies/2026-10-27-sdlc/README.md Resolved
Comment thread policies/2026-10-27-sdlc/README.md Outdated
### 3.6 Least Functionality & Software Execution Control (NIST SP 800-171 Rev 2: 3.4.6, 3.4.7, 3.4.8)

* **Permit-by-exception for images (3.4.8):** EKS and OpenShift refuse to run unsigned or unverified container images (Section 6.1).
* **GitHub Actions:** Third-party actions are pinned to a full commit SHA and restricted to an approved list at the organization level. [TBD: confirm.]

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Not currently, but we can do that. This can be enforced at the organizational level, there is a "Require actions to be pinned to a full-length commit SHA" setting.

We currently rely on dependabot to get notifications about new releases to github actions. My brief reading is that we can continue to rely on dependabot for update notifications (as long as we follow specific syntax in our workflows), but we may lose out on security notications (https://docs.github.com/en/code-security/concepts/supply-chain-security/dependabot-alerts).

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can we remove the TBD mark here? We know what the current situation looks like, and the new actions-lock extension looks like it gives us a solid path forward meeting this requirement.

Comment thread policies/2026-10-27-sdlc/README.md Outdated

### 3.7 Software Inventory & Dependency Management (NIST SP 800-171 Rev 2: 3.4.1)

* **Python:** Dependencies are pinned to explicit versions with hashes (`pip install --require-hashes`) in requirements files. [TBD: confirm.]

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pinned to a specific crytpographic hash by the uv.lock package lockfile.

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Updated.

Comment thread policies/2026-10-27-sdlc/README.md Resolved
Comment thread policies/2026-10-27-sdlc/README.md Resolved

### 3.6 Least Functionality & Software Execution Control (NIST SP 800-171 Rev 2: 3.4.6, 3.4.7, 3.4.8)

* **Permit-by-exception for images (3.4.8):** EKS and OpenShift refuse to run unsigned or unverified container images (Section 6.1).

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Just to get some clarification on the scope of this one. So, for an openshift cluster I imagine we'll trust the signing key of a vendor (nvidia, everpure) so that we can run their container images. What about tenant workloads?

@knikolla knikolla Oct 30, 2026 •

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

In my understanding, tenant workloads are the customers responsibility. And they'd need to have similar processes in place since they would be running code touching CUI, thus subject to 800-171.

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

so does this become part of onboaring a customer to a multitenant openshift cluster where we setup some way to trust their images to run and only into their namespace.

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes?

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This might require reviving he internal Quay server for NIST/HIPAA environment MOC uploads that's been discussed before. It also comes with Clair for analyzing container images. This is not planned for MOC 1.5.

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It still needs to be defined in the policy, and then the gap between reality (1.5) and the policy written in the self-assessment.


* **Python:** Dependencies are pinned to explicit versions with hashes (`pip install --require-hashes`) in requirements files. [TBD: confirm.]
* **Terraform/OpenTofu:** Provider and module versions are pinned, and dependency lock files (`.terraform.lock.hcl`) are committed.
* **Containers and operators:** Base images are referenced by digest. Operators and Helm charts are pinned to specific versions.

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is this suggesting pinning 3rd party operators and helm charts (like nvidia, everpure) to specific version OR only the operators/charts that we have developed? Asking because I am trying to understand why this policy about SDLC call out this operational thing. I am not against it, but just want to understand the scope.

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Including 3rd party. The reasoning is malicious releases caused by taking over the release infrastructure of an open source project. Pinning to a specific hash/digest ensures that someone compromising a specific base image or github action cannot cut a new release (or rewrite old release tags) and our tooling would automatically pick it up. See https://snyk.io/blog/reconstructing-tj-actions-changed-files-github-actions-compromise/ for an example of such a situation.

Comment thread policies/2026-10-27-sdlc/README.md Outdated

* **Key rotation:** Customer-managed cryptographic keys (AWS KMS) are rotated automatically at least every **90 days**.
* **Separation of duties:** Individuals who administer keys (create, change policy, schedule deletion) are not the same individuals who use keys to encrypt or decrypt data, and no single individual can both change a key policy and use the key. Key policies are managed in code and changed only through the process in Section 3.3.
* **Certificates:** TLS/SSL certificates are provisioned and managed via **AWS Certificate Manager (ACM)** or automated via **Let's Encrypt**, with automated renewal.

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We want to setup a private CA to sign certificates for our storage appliance (for stuff like NFS over TLS). That would mean the certificates used by it are neither of the ones you mentioned.

AWS does have private CA, but its expensive to use given how few certificates we'll be signing with it.

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Good point, I'll add something about our own self-signed certs in this. Thanks for the feedback.

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

...and also for BMC/other device access. I just posted the local CA ADR to #52.


### 1.1 Scope & Applicability

This Secure Software Development Life Cycle (SDLC) policy is a supporting document required by, and referenced from, the System Security Plan (SSP). It is not an assessment of all NIST SP 800-171 requirements. It addresses only the requirements that are touched by writing and running code in production within the CUI boundary. How the system implements each requirement (components, settings, tools, and evidence) is documented in the SSP.

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The "running code in production" seems to vastly increase the scope of this document. I guess that's why this policy goes into the details of what SSL certs to use, and to pin operator versions etc? Not a complaint but I am also just trying to understand.

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

All code that runs within the CUI boundary needs to be included in this. And the infrastructure as code runs within the CUI boundary, so should be treated as code per the SDLC.

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We expect to also offer The Open Accelerator (TOA) and Open Science Clusters, which sit outside the CUI boundary.

Two questions:

  1. Scope: As written, 1.1 limits this policy to code running within the CUI boundary, so it would not cover TOA or Open Science. Should 1.1 say that explicitly, and say what (if anything) governs code outside the boundary?
  2. Structure: If policies and ADRs will differ by environment type, options that come to mind:
  3. Separate repo for NIST/HIPAA-regulated work vs. Open Science.
    • Pros: we may end up with NIST 800-53 or customer-specific requirements for each, and this would support that.
    • Cons: if a control is not required for a given environment, we are less likely to prioritize it there.
  4. One repo, with each document declaring which environments it applies to. larsks's point suggests this belongs in frontmatter, not in the body, but I think we may have situations where it may need to be tied to an aspect of the policy or adr.
    • Pros: more likely to be accomplished for multiple cluster types.
    • Cons: increased complexity; could muddy the waters for audits.
  5. Some other method?

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

  1. That is correct. This only targets code running within the CUI boundary.

I'm not sure about the best way to achieve what you are proposing.

@hpdempsey hpdempsey Nov 3, 2026 •

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

In our call with Davic C. today he said that project data is the only thing that could be defined as CUI in an MOC cluster. So is the CUI boundary defined by any running code that can access the project namespace or resources associated with that namespace for processing, memory or remote storage? Or do you have a different definition? I see there are several things you are pointing out are NOT CUI, but should we assume that everything not excluded shoudl be considered CUI?

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The CMMC scoping guide is a good way to find out what the boundary is. https://dodcio.defense.gov/Portals/0/Documents/CMMC/ScopingGuideL2.pdf

So is the CUI boundary defined by any running code that can access the project namespace or resources associated with that namespace for processing, memory or remote storage? Or do you have a different definition? I see there are several things you are pointing out are NOT CUI, but should we assume that everything not excluded shoudl be considered CUI?

Yes. It's not just about the CUI itself. The scoping guide defines as in-scope:

  • Assets that process, store, or transmit CUI
  • Assets that provide security functions or capabilities
  • Assets that can, but are not intended to, process, store, or transmit CUI

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The CMMC scoping guide is a good way to find out what the boundary is. https://dodcio.defense.gov/Portals/0/Documents/CMMC/ScopingGuideL2.pdf

So is the CUI boundary defined by any running code that can access the project namespace or resources associated with that namespace for processing, memory or remote storage? Or do you have a different definition? I see there are several things you are pointing out are NOT CUI, but should we assume that everything not excluded shoudl be considered CUI?

Yes. It's not just about the CUI itself. The scoping guide defines as in-scope:

  • Assets that process, store, or transmit CUI
  • Assets that provide security functions or capabilities
  • Assets that can, but are not intended to, process, store, or transmit CUI

Thanks, that seems like a good working definition. Perhaps add this to the policy, even if it is just in an appendix, because not everyone who needs to follow this SDLC policy to produce code for the MOC will be aware of that definition. We are likely to get more Red Hat folks helping prepare for the push towards production deployments.

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Added link to CMMC scoping guide.

@msdisme

msdisme commented Oct 30, 2026

Copy link
Copy Markdown
Contributor

@naved001 is adding comments, I will be the MOC approver. I plan to work on it after 11:30


### 1.1 Scope & Applicability

This Secure Software Development Life Cycle (SDLC) policy is a supporting document required by, and referenced from, the System Security Plan (SSP). It is not an assessment of all NIST SP 800-171 requirements. It addresses only the requirements that are touched by writing and running code in production within the CUI boundary. How the system implements each requirement (components, settings, tools, and evidence) is documented in the SSP.

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We expect to also offer The Open Accelerator (TOA) and Open Science Clusters, which sit outside the CUI boundary.

Two questions:

  1. Scope: As written, 1.1 limits this policy to code running within the CUI boundary, so it would not cover TOA or Open Science. Should 1.1 say that explicitly, and say what (if anything) governs code outside the boundary?
  2. Structure: If policies and ADRs will differ by environment type, options that come to mind:
  3. Separate repo for NIST/HIPAA-regulated work vs. Open Science.
    • Pros: we may end up with NIST 800-53 or customer-specific requirements for each, and this would support that.
    • Cons: if a control is not required for a given environment, we are less likely to prioritize it there.
  4. One repo, with each document declaring which environments it applies to. larsks's point suggests this belongs in frontmatter, not in the body, but I think we may have situations where it may need to be tied to an aspect of the policy or adr.
    • Pros: more likely to be accomplished for multiple cluster types.
    • Cons: increased complexity; could muddy the waters for audits.
  5. Some other method?

Comment thread policies/2026-10-27-sdlc/README.md Resolved
Comment thread policies/2026-10-27-sdlc/README.md Outdated

### 4.1 Secure Coding Standards (NIST SP 800-171 Rev 2: 3.13.2)

* Python development teams must adhere to the **OpenSSF Secure Coding One Stop Shop for Python** guidelines.

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Proposed Replacment for 4.1:

  • Python code must follow the OpenSSF Secure Coding One Stop Shop for Python guide (pyscg). Rules that static analysis can detect are enforced as required checks in CI, using the tools and mapping maintained in [location TBD]. Remaining rules are verified through pull request review by a designated security reviewer (Section 3.3). [TBD: perhaps the CI scanning belongs in Section 4.2 (Automated Security Scanning), with this section keeping only the standard and the review requirement.]
  • The mapping of each pyscg rule to an automated check or to review is maintained in the repository, so coverage and gaps are visible. [TBD: tool selection per Section 4.2.]

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm hesitant to say that we will implement the pyscg with automated scanning. We can revise this section if required later.

Comment thread policies/2026-10-27-sdlc/README.md Outdated
4. Drift detection (3.2); emergency change process (3.4); change access restrictions (3.5).
5. Confirm Actions pinning and allow-list, and minimal images (3.6); confirm Python hashes; Keycloak and LDAP dependency controls (3.7).
6. SAST tool selection; secret scanning confirmation; rescan frequency (4.2).
7. Confirm the mapping of BU Moderate to CVSS Medium, and KEV handling (4.3).

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is not meant to block approval, rather to identify that we have a variety of answers to this question and to gather it together for review with David C. and Brian.

Comparing the §4.3 values against David Corbett's Continuous Monitoring Plan (AWS), including his comments, and the MOC 2.0 Compliance Operations Guide. The BU columns come from the Cyber Risk Management Standard and the Vulnerability Management Standard.

PR 50 §4.3 (CUI systems / non-CUI) David's AWS plan Scott's Ops Guide BU Cyber Risk Mgmt Standard (CUI / non-CUI)
Critical 5 / 14 days 15 calendar days 10 business days (labelled FedRAMP) 5 days / 2 weeks
High 14 / 28 days 30 days 30 days 2 weeks / 4 weeks
Medium or Moderate 28 / 60 days 60 days in §7, but 90 in the §6.1 Inspector row 60 days (Moderate/Low together) 4 weeks / 90 days
Low 60 / 60 days 60 days 60 days 90 / 180 days
Overall cap 60 days for everything David's comment: "BU has an outward limit of 60 days" 60 days, links the BU standard 60 days (Vulnerability Management Standard)
Goes to POA&M After missing the deadline, as an exception tracked under §7.3 Cannot fix in 90 days (§7). Items open over 90 days are reported (§10) Cannot fix in 90 days. POA&M items must be fully remediated within 180 days n/a
Clock starts Identified, or Risk Owner notified Date of detection Not stated "Becoming known"

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

policies and procedures, privacy notices, security risk analyses, incident reports, and complaint logs—for 6 years

  • and what evidence is needed for future audits

IS&T - Servicenow allows us to not need to store since the details will be stored in the issue.

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

feedback from David C: CUI is about patient data not about info in logs:
CUI is sensitive information that does not meet the criteria for classification but must still be protected. I

security logs for one year

all operational and compliance activities documentation kept for 6 yars

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

With respect to 4.3 .. do we have to consider long running jobs?
For example:
A job starts running on d=1 and is expected to last 10 days (d=11).
On d=2 a critical vulnerability is published and our target to remediation is 5 days (d=7).
What happens to that running job? Do we interrupt the job or let it ride until completion and remediate (for whatever that looks like) when it's done.
If it's the later (we let the job run) do we need to adjust language, or is the risk acceptance procedure the way to mitigate?

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't believe the customer would want to pay the usage charges for keeping resources allcoated to a long-running job that couldn't proceed. They would probably want a credit for resources used on the interrupted job and then they would have to restart after the issue was resolved.

@bu-renm

bu-renm commented Nov 2, 2026

Copy link
Copy Markdown
Contributor

In the same vein as as my comment here, can anyone find the specific controls that mandate we don't deploy automatically to prod?

My frame is that the PR approval process is the point at which changes "get made" to prod, not when ArgoCD syncs.

I'm using latest 800-53 directly here, but:

CM-03(01):

 b. Notify [Assignment: organization-defined approval authorities] of proposed changes to the system and request change approval;
...
d. Prohibit changes to the system until designated approvals are received;

Sure, and then:

CM-03(03):

Control Statement

Implement changes to the current system baseline and deploy the updated baseline across the installed base using [Assignment: organization-defined automated mechanisms].

Discussion:

Automated tools can improve the accuracy, consistency, and availability of configuration baseline information. Automation can also provide data aggregation and data correlation capabilities, alerting mechanisms, and dashboards to support risk-based decision-making within the organization.

This seems to endorse the "CD" part of CICD using something like ArgoCD.

There's:

CM-05(04):

Control Statement

Enforce dual authorization for implementing changes to [Assignment: organization-defined system components and system-level information].

Discussion:

Organizations employ dual authorization to help ensure that any changes to selected system components and information cannot occur unless two qualified individuals approve and implement such changes. The two individuals possess the skills and expertise to determine if the proposed changes are correct implementations of approved changes. The individuals are also accountable for the changes. Dual authorization may also be known as two-person control. To reduce the risk of collusion, organizations consider rotating dual authorization duties to other individuals. System-level information includes operational procedures.

Which seems to map to the Github PR model we're already using.

And also:

CM-05(05):

...
a. Limit privileges to change system components and system-related information within a production or operational environment; and
...

Which tells us to treat prod with particular care.

But I don't see anything that specifically says we can't use a tool that automatically reconciles source control with prod state. Especially with CM-03(03), my reading is that the PR workflow takes care of the gating for changes in a NIST environment.

I'd love to be proven wrong here, but I'd equally love if we could have consistent tooling across NIST/non-NIST.

That being said, it's probably good to have e.g. a separate Github Enterprise account to narrow the compliance boundary and to explore self-hosted runners for the same reason. But I'm not sure we need to throw out ArgoCD with this bathwater.

@knikolla

knikolla commented Nov 2, 2026

Copy link
Copy Markdown
Author

In the same vein as as my comment here, can anyone find the specific controls that mandate we don't deploy automatically to prod?

My frame is that the PR approval process is the point at which changes "get made" to prod, not when ArgoCD syncs.

I'm using latest 800-53 directly here, but:

CM-03(01):

 b. Notify [Assignment: organization-defined approval authorities] of proposed changes to the system and request change approval;
...
d. Prohibit changes to the system until designated approvals are received;

Sure, and then:

CM-03(03):

Control Statement

Implement changes to the current system baseline and deploy the updated baseline across the installed base using [Assignment: organization-defined automated mechanisms].

Discussion:

Automated tools can improve the accuracy, consistency, and availability of configuration baseline information. Automation can also provide data aggregation and data correlation capabilities, alerting mechanisms, and dashboards to support risk-based decision-making within the organization.

This seems to endorse the "CD" part of CICD using something like ArgoCD.

There's:

CM-05(04):

Control Statement

Enforce dual authorization for implementing changes to [Assignment: organization-defined system components and system-level information].

Discussion:

Organizations employ dual authorization to help ensure that any changes to selected system components and information cannot occur unless two qualified individuals approve and implement such changes. The two individuals possess the skills and expertise to determine if the proposed changes are correct implementations of approved changes. The individuals are also accountable for the changes. Dual authorization may also be known as two-person control. To reduce the risk of collusion, organizations consider rotating dual authorization duties to other individuals. System-level information includes operational procedures.

Which seems to map to the Github PR model we're already using.

And also:

CM-05(05):

...
a. Limit privileges to change system components and system-related information within a production or operational environment; and
...

Which tells us to treat prod with particular care.

But I don't see anything that specifically says we can't use a tool that automatically reconciles source control with prod state. Especially with CM-03(03), my reading is that the PR workflow takes care of the gating for changes in a NIST environment.

I'd love to be proven wrong here, but I'd equally love if we could have consistent tooling across NIST/non-NIST.

That being said, it's probably good to have e.g. a separate Github Enterprise account to narrow the compliance boundary and to explore self-hosted runners for the same reason. But I'm not sure we need to throw out ArgoCD with this bathwater.

I don't think there is any control that forbids deploying automatically to prod via something like ArgoCD. The only limitation that I'm aware is that whatever runs the tooling that has power to change the config of prod is part of the scope, therefore it can't run in GitHub actions that are hosted by GitHub. So as long as we either run ArgoCD and/or the GitHub self-hosted runners we should be fine.

@hpdempsey

hpdempsey commented Nov 4, 2026 •

Copy link
Copy Markdown

In the same vein as as my comment here, can anyone find the specific controls that mandate we don't deploy automatically to prod?

This is a long-standing MOC policy, not a NIST/HIPAA control as far as I know. I think Lars is considering one exception, which would be when scanning detects production deployed code that is different from the intended deployment record and reverses it automtically in the NIST/HIPAA compliant clluster. Even that is controversial for some folks in Red Hat.

@hpdempsey

Copy link
Copy Markdown

@knikolla there are outstanding requests for changes from @larsks and @msdisme to be made or resolved another way. Can you please follow up so we can proceed with approvals?

@larsks larsks left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I've left some new comments, and I have un-resolved some old ones that seemed to suggest document changes that hadn't been made.

There are a lot of TBD markers left in this document; while I think we may be able to resolve several of them, is the goal to eliminate them all before merging this document?

Comment thread policies/2026-10-27-sdlc/README.md Outdated
### 1.2 CUI Boundary, Deployment Model & External Systems

* **Platform:** The platform consists of an on-premises OpenShift cluster and foundational services (for example Keycloak and LDAP) on Amazon EKS in AWS. Both are within the CUI boundary. The SSP describes the architecture, the boundary, and the network path between them.
* **Deployment Model:** CI/CD does not deploy to production. Jobs running inside the CUI boundary, pull approved configuration from the public and private GitHub repositories and apply it to production, including the on-premises OpenShift cluster (see Section 6.1).

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Remove an extraneous comma:

Suggested change
* **Deployment Model:** CI/CD does not deploy to production. Jobs running inside the CUI boundary, pull approved configuration from the public and private GitHub repositories and apply it to production, including the on-premises OpenShift cluster (see Section 6.1).
* **Deployment Model:** CI/CD does not deploy to production. Jobs running inside the CUI boundary pull approved configuration from the public and private GitHub repositories and apply it to production, including the on-premises OpenShift cluster (see Section 6.1).

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Removed the comma

Comment thread policies/2026-10-27-sdlc/README.md Outdated

* **Platform:** The platform consists of an on-premises OpenShift cluster and foundational services (for example Keycloak and LDAP) on Amazon EKS in AWS. Both are within the CUI boundary. The SSP describes the architecture, the boundary, and the network path between them.
* **Deployment Model:** CI/CD does not deploy to production. Jobs running inside the CUI boundary, pull approved configuration from the public and private GitHub repositories and apply it to production, including the on-premises OpenShift cluster (see Section 6.1).
* **GitHub as an External System:** GitHub (including GitHub Actions) is not an in-scope external system for the purposes of NIST SP 800-171 3.1.20. It holds no authorization to change production and no credentials for production or for anything inside the CUI boundary. [TBD: confirm.] Controls on what reaches the branches that are pulled into production are described in Sections 3 and 6.1.

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We've confirmed what our GitHub configuration looks like right now. Do we still need the TBD mark here? I would like to see an answer to @msdisme's question regarding on-premise runners -- is that an acceptable solution? I believe (but have not confirmed) that we are still granting GitHub access to credentials in that case.

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Removed the TBD mark.

Comment thread policies/2026-10-27-sdlc/README.md Outdated
### 3.1 Baselines & Source Control (NIST SP 800-171 Rev 2: 3.4.1)

* **Version Control System:** Git is used for all code, scripts, and infrastructure configuration files. Configuration in Git is the baseline for production.
* **Branch Protection Policies:** Direct commits or force pushes to the primary production branch (main) are disabled.

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You gave me a thumbs up here, but I don't see any related change to the document. Are we going to leave the phrasing as-is, or update it to cover this use case?

Comment thread policies/2026-10-27-sdlc/README.md Outdated
### 3.6 Least Functionality & Software Execution Control (NIST SP 800-171 Rev 2: 3.4.6, 3.4.7, 3.4.8)

* **Permit-by-exception for images (3.4.8):** EKS and OpenShift refuse to run unsigned or unverified container images (Section 6.1).
* **GitHub Actions:** Third-party actions are pinned to a full commit SHA and restricted to an approved list at the organization level. [TBD: confirm.]

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can we remove the TBD mark here? We know what the current situation looks like, and the new actions-lock extension looks like it gives us a solid path forward meeting this requirement.

Comment thread policies/2026-10-27-sdlc/README.md Outdated

* **Permit-by-exception for images (3.4.8):** EKS and OpenShift refuse to run unsigned or unverified container images (Section 6.1).
* **GitHub Actions:** Third-party actions are pinned to a full commit SHA and restricted to an approved list at the organization level. [TBD: confirm.]
* **Images (3.4.6, 3.4.7):** Container images are built from minimal base images and expose only the ports, services, and protocols they require. [TBD: confirm.]

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What needs to be confirmed here -- that this is a requirement, or that this is how we are currently building images?

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I was going to confirm which base image we use, but since that probably vaies I'm just dropping the TBD for now.


### 5.2 Key Management (NIST SP 800-171 Rev 2: 3.13.10, 3.1.4)

* **Key rotation:** Customer-managed cryptographic keys (AWS KMS) are rotated automatically at least every **90 days**.

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Who is the "customer" in this statement? Is this setting a requirement on how we manage secrets, or on how our customers manage secrets?

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Comment thread policies/2026-10-27-sdlc/README.md Outdated

* **Key rotation:** Customer-managed cryptographic keys (AWS KMS) are rotated automatically at least every **90 days**.
* **Separation of duties:** Individuals who administer keys (create, change policy, schedule deletion) are not the same individuals who use keys to encrypt or decrypt data, and no single individual can both change a key policy and use the key. Key policies are managed in code and changed only through the process in Section 3.3.
* **Certificates:** TLS/SSL certificates are provisioned and managed via **AWS Certificate Manager (ACM)** or automated via **Let's Encrypt**, with automated renewal. For systems for which we cannot get public certificates we deploy and manage self-signed certificates. [TBD: link to ADR for self-signed certificates once it merges]

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Added link.

Comment thread policies/2026-10-27-sdlc/README.md Outdated
* Container images are built automatically via GitHub Actions upon publishing a formal GitHub release tag.
* Container images are cryptographically signed using **Sigstore** to guarantee supply chain integrity and provenance.
* **EKS and OpenShift refuse to run unsigned or unverified images.** [TBD: the trusted signing identity, and how it is rotated or revoked if compromised.] The SSP documents the enforcement mechanism on each platform.
* **Pull-based deployment:** CI/CD does not deploy to production. Jobs running inside the CUI boundary pull the approved configuration from the protected branches of the public and private repositories and apply it to production, including the on-premises OpenShift cluster. [TBD: confirm how Terraform/OpenTofu changes are applied, and whether the jobs verify commit or tag signatures.]

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Earlier you said we didn't need to require signed commits, but here you are referring to commit and tag signatures. I am confused!

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't know whether we need or not signed commits, hence the TBD. A question David C would have answered by reviewing the SDLC.


### 6.2 Audit Trail (NIST SP 800-171 Rev 2: 3.3.1, 3.3.2)

* Changes to the platform are traceable to authenticated individuals through GitHub commit history, pull request reviews and approvals, release tags, and AWS CloudTrail.

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

...through GitHub commit history...

Noting that without signed commits we cannot reliably use the commit history to identify individuals who made changes.

GIT_AUTHOR_NAME='Kristi Nikolla' \
GIT_AUTHOR_EMAIL='knikolla@bu.edu' \
GIT_COMMITTER_NAME='Kristi Nikolla' \
GIT_COMMITTER_EMAIL='knikolla@bu.edu' \
git commit -m "I haxored your server"

Which gives us:

$ git cat-file -p HEAD
tree f779bce6dc1206c30d21d6e5c92cfb87d822730e
author Kristi Nikolla <knikolla@bu.edu> 1795014766 -0500
committer Kristi Nikolla <knikolla@bu.edu> 1795014766 -0500

I haxored your server

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'll revise the signed commits aspect once we have clarity over the process there.

### 6.2 Audit Trail (NIST SP 800-171 Rev 2: 3.3.1, 3.3.2)

* Changes to the platform are traceable to authenticated individuals through GitHub commit history, pull request reviews and approvals, release tags, and AWS CloudTrail.
* **GitHub audit log:** Because GitHub's own retention is limited, the GitHub organization audit log is streamed to retained storage. [TBD: destination.]

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think this will require that we are using one of GitHub's Enterprise tiers (either self-hosted GitHub Enterprise Server, or GitHub Enterprise Cloud), based on https://docs.github.com/en/enterprise-cloud@latest/admin/monitoring-activity-in-your-enterprise/reviewing-audit-logs-for-your-enterprise/streaming-the-audit-log-for-your-enterprise.

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Adding link to CCI-MOC/MOC-issues#709

- Added CMMC scoping guide link.
- Updated deployment model to mention argocd and on-prem runners.
- Usage of GitHub teams to manage access.
- That GitHub doesn't permit access to secrets from forks.
- Removed the TBD of secret scanning, and a few other TBDs.
- Made the security training more general, hopefully.
- Added links to self-signed CA and drift detection docs ADRs.
- Reduced open items list.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

ops-policy This is a policy document

Projects

None yet

Development

Successfully merging this pull request may close these issues.

7 participants