SOC 2 and open source dependencies: is dependency scanning or an SBOM required?
Updated
No. None of the sources fetched for this page says the SOC 2 criteria name a dependency scanner or a software bill of materials (SBOM). Vulert, a company that sells a dependency scanner, writes that SOC 2 "does not usually say, 'You must use this exact dependency scanner.'" What an auditor tests is whether you find known vulnerabilities in the packages you ship, fix or accept them on a timeline, and follow the rule you wrote. An open alert that is older than your own deadline is the usual problem, not the missing tool.
What the sources say
Most pages on this question are written by companies that sell scanners or SBOM tools. Three pages were fetched for this guide: two vendors and one assessment firm. None says the criteria require either.
| Source | Who they are | What it says |
|---|---|---|
| Vulert (Dawood Ikhlaq, co-founder, 30 April 2026) | Sells a dependency scanner | SOC 2 "evaluates whether your controls are designed and operating effectively". On SBOMs: "an SBOM is only an inventory" |
| GitLab (Fernando Diaz, 22 January 2025) | Sells a source control and CI/CD platform | Lists "Dependency Scanning" among its scanners and says it "can generate a Software Bill of Materials (SBOM)". Names no criterion for either |
| Schellman (Sully Perella, senior manager; 2 October 2024, updated 20 March 2026) | Assessment firm that performs SOC examinations and PCI DSS validations | Describes an SBOM as "a software inventory" and names contracts, PCI and a US executive order as where one is required. The article does not mention SOC 2 |
Sources: (Vulert) (GitLab) (Schellman). Two limits. Schellman's article is filed under payment card assessments and is about SBOMs in general, so its silence on SOC 2 is weaker evidence than a direct statement would be. And no accounting firm's page was found that answers this question head-on; the searches for this guide returned scanner sellers. Your CPA firm has the last word for your system.
Where dependencies land in the criteria
Vulert ties dependencies to two criteria. This is one vendor's mapping, and it adds that an auditor "may also connect dependency security to risk assessment, vendor management, incident response, or access controls" (Vulert).
| Criterion | What Vulert says it means for dependencies | What you would show |
|---|---|---|
| CC7.1 | "your team should have a defined way to identify newly discovered vulnerabilities affecting the packages your software already uses" | Scan results and alerts with dates across the whole period, and who reviewed them |
| CC8.1 | "Dependency updates are software changes." An upgrade "should be authorized, tested, approved, and documented before deployment" | The pull request for each fix, with its review, its checks and the deploy record |
The right column is this page's reading, not a list from a source. The CC7.1 point has a practical consequence that Vulert spells out: "A vulnerability disclosed today can affect code you shipped six months ago." A scan that runs only when you release does not see it. Scan timing in general is covered in how often to run vulnerability scans, and the change path in SOC 2 change management.
The trap: your own deadlines
Vulert's sharpest line is about timing: "A single vulnerability scan performed the week before your audit is not evidence of continuous monitoring" (Vulert). It sells continuous monitoring, so read the emphasis as a seller's. The mechanism is real, though: a Type II examination covers a period, and the auditor can ask for results from any part of it.
The second half of the trap is the written rule. Vulert prints example fix timelines and then says: "These timelines are not universal SOC2 rules, but they give auditors something concrete to test." That cuts both ways. Once a deadline is in your policy, every alert has a due date, and an alert list with old open items is a list of misses against your own rule. Policyseed sells a kit, so read the text below as one example of such a rule. It is rendered from the policy templates for the Northwind Cloud Inc sample intake, not typed by hand, so it changes when a template does.
The deadlines (Vulnerability and Patch Management Policy, 4.3)
Findings are remediated within timelines measured from detection: Critical within 7 days, High within 30 days, Medium within 90 days, and Low within 180 days or at the next scheduled maintenance, whichever is sooner. Where no fix exists for a Critical finding in an internet-facing system, a mitigating control such as a firewall rule, feature disablement or configuration change is applied within 72 hours.
Dependency updates (Secure Software Development Policy, 5.4)
Dependency management. Automated update pull requests are reviewed weekly. Security updates merge within the Vulnerability and Patch Management Policy timelines: Critical within 7 days, High within 30 days, Medium within 90 days. Each quarter Engineering reviews the dependency inventory for unmaintained or end-of-life components and plans replacements.
When a fix cannot land on time (Vulnerability and Patch Management Policy, 4.6)
Where a finding cannot be remediated on time, the owner submits an exception with the reason, compensating controls and planned remediation date. The Security Owner approves exceptions for Medium and Low findings; Critical and High exceptions also require Executive Management approval. Exceptions last no more than 90 days and are re-reviewed before renewal.
If your team cannot review update pull requests every week or fix a Critical finding in the stated number of days, change the numbers before the policy is approved. A longer deadline that is met is easier to defend than a short one that is missed, because the written rule is what gets tested.
What to keep
Vulert's summary of the evidence: "Keep scan reports, alert logs, Jira history, pull requests, test records, deployment logs, and exception approvals" (Vulert). For a small team on one repository host that comes down to four things:
- Scan results across the whole period. The CI run history that shows the dependency check ran on each pull request, plus the scheduled scan or alert feed. Check that your host keeps that history for as long as your observation period.
- One ticket or pull request per fix. With the date the finding was detected and the date the fix merged, so the time to fix can be read against the deadline. The change request form covers changes that do not go through a pull request.
- A record for what stays open. Vulert asks for exceptions with "documented approval, business justification, compensating controls, and expiration dates". The policy exception register has those columns.
- The dependency manifest per release. The lockfile or SBOM that was built into each production release, kept with the release record.
How many of these an auditor pulls is covered in SOC 2 sampling. One limit of this site's own tooling: the open-source Policyseed CLI's policyseed evidence command (see integrations) records whether Dependabot alerts are switched on for a GitHub repository. It does not export the alerts or their dates, so the tickets and pull requests above still have to come from you.
The SBOM question
Schellman's definition: "Think of an SBOM as a software inventory", one that "identifies all the components and dependencies used by your software, including the specific version(s) of each" (Schellman). Vulert agrees it helps and adds the limit: "an SBOM is only an inventory. For SOC2 vulnerability management, teams still need to analyze the SBOM for known vulnerabilities, monitor new CVEs, and document remediation activity." Its own checklist accepts either form: each in-scope application "has identified dependency files or SBOM files", and it gives lockfiles such as package-lock.json and go.sum as examples of dependency files (Vulert).
So by these sources a formal SBOM file is not what SOC 2 asks for. The generated policies go further than the sources do, and a company that adopts them has made two promises of its own:
The inventory (Asset Management Policy, 4.6)
Engineering maintains a software bill of materials for Northwind listing the languages, frameworks, libraries and container base images in use and their versions, generated automatically from GitHub and GitHub Actions wherever possible, so that vulnerability notices can be matched to what is actually deployed. Open-source components are used within their license terms, and the IT/Operations Lead keeps license and subscription records.
When it is produced (Asset Management Policy, 5.6)
Software bill of materials. The dependency manifest for Northwind is generated by GitHub Actions on each production build. Engineering stores the latest manifest with the release record, and the Vulnerability and Patch Management Policy uses it to match advisories to deployed components.
Item 5.6 says a manifest is generated on each production build and stored with the release record. An auditor who reads that sentence can ask for the manifest of a release picked from the middle of the period. If your build does not store one, either add the step or edit the sentence. The asset inventory workbook covers devices, cloud resources and SaaS; the dependency manifest lives with the code, not in that sheet.
What the generated policies say about scanning
The generator writes dependency scanning into three policies. None of these depends on an intake answer, so every adopter gets them.
On every pull request (Change Management Policy, 5.3)
Automated checks. Every pull request must pass unit and integration tests, linting, dependency vulnerability scanning, static application security testing, secret scanning, infrastructure-as-code validation and a successful build in GitHub Actions. Suppressing a failing check requires written justification in the pull request and Security Owner approval.
The sources of findings (Vulnerability and Patch Management Policy, 4.1)
Northwind Cloud Inc maintains continuous discovery across in-scope assets from at least these sources: software composition and static analysis in GitHub Actions on every pull request and on a schedule; secret scanning in GitHub; container image scanning before deployment and on a recurring schedule; the native security posture service of each platform in AWS and Vercel; endpoint patch and compliance reporting from Kandji; an annual independent penetration test; vendor advisories; published threat intelligence including the CISA Known Exploited Vulnerabilities catalog; and reports received through the disclosure channel.
Checking that the scanners still run (Vulnerability and Patch Management Policy, 5.1)
Continuous scanning. Engineering configures each source in 4.1 to run automatically and route findings into the register or a designated alert channel, and confirms weekly that every source produced results in the past seven days. A source that stops reporting is a High finding against the scanning system itself.
Where dependencies come from (Secure Software Development Policy, 4.6)
Dependencies come only from official registries, are pinned with lockfiles, are kept current through automated update pull requests, and are not used when end-of-life, unmaintained or incompatibly licensed. Dependencies with known Critical or High vulnerabilities are updated within the Vulnerability and Patch Management Policy timelines.
Three things to notice. First, the pull request rule is absolute: a repository without a dependency check in its pipeline does not match it. Second, item 5.1 of the Vulnerability and Patch Management Policy asks for a weekly confirmation that every source produced results, which is a recurring task someone has to own. Third, suppressing a failing check needs a written reason and the Security Owner's approval, so a habit of dismissing alerts without a note is itself a departure from the policy.
Other frameworks and customer contracts
The requirement people half-remember usually comes from somewhere else. Schellman: "Whether it's written into your contracts or part of your compliance with PCI or the Executive Order on Improving the Nation's Cybersecurity, all these concepts require organizations to have an inventory of what comprises their software" (Schellman).
For PCI, the guidance site WithPCI quotes requirement 6.3.2 of PCI DSS v4.0.1: "An inventory of bespoke and custom software, and third-party software components incorporated into bespoke and custom software is maintained to facilitate vulnerability and patch management", with a note that it became required after 31 March 2025 (WithPCI). That is a secondary site quoting the standard; the standard itself sits behind a license page and was not read for this guide. It applies only if card data is in your scope. The executive order is named here on Schellman's word alone; its text was not fetched, so this page says nothing about whom it binds.
A routine a small team can keep
This list is the page's own suggestion, built from the items above, not a list from a source.
- Turn on one detection source per repository and make the dependency check a required check on pull requests.
- Triage on a fixed day. Each new finding gets a severity, an owner and a due date from your policy's timeline.
- Fix through the normal pull request. No separate path for security updates; the review and the passing checks are the evidence.
- Write down what you will not fix yet in the exception register, with a compensating control and an end date.
- Store the manifest with each release. The lockfile at the release commit, or a generated SBOM file attached to the release.
- Report the counts monthly. Open findings by severity, fixed on time and overdue. The security metrics report has the layout.
Frequently asked questions
- Does SOC 2 require dependency scanning?
- Not by name, according to the sources fetched for this page. Vulert, which sells a dependency scanner, writes that SOC 2 "does not usually say, 'You must use this exact dependency scanner.'" and that it "evaluates whether your controls are designed and operating effectively". GitLab's SOC 2 guide lists dependency scanning among its security scanners without tying it to a criterion. What gets tested is whether you find known vulnerabilities in what you run, act on them, and follow your own written rule.
- Does SOC 2 require an SBOM?
- No fetched source says so. Vulert calls an SBOM "only an inventory" and its checklist accepts "dependency files or SBOM files". Schellman, a firm that performs SOC examinations and PCI assessments, names PCI and a US executive order as the places an inventory of software components is required, and its article does not mention SOC 2. One caution: the generated Asset Management Policy promises a dependency manifest on each production build, so a company that adopts it has made the promise itself.
- Are Dependabot alerts enough?
- None of the fetched sources discusses Dependabot in its text, so this is the page's reasoning, not a citation. An alert feed is a way to detect. Vulert's list of what auditors look for goes past detection: alerts reviewed by a named person, tickets with owners and due dates, the pull request that fixed each one, and a record for anything accepted instead of fixed. Alerts that are switched on and never triaged show the opposite of what you want: a dated list of findings that stayed open past your own deadlines.
- What about dev-only dependencies?
- No fetched source covers them. The generated Vulnerability and Patch Management Policy rates a finding from its CVSS score and then adjusts it for context, including whether it is reachable in the product, and the Security Owner may change a rating with a recorded justification. So a build-time-only package can be rated lower, but the adjustment has to be written down. Closing the alert without a note leaves no evidence of that judgment.
- Do we need to give customers an SBOM?
- SOC 2 is not where that comes from, by these sources. Schellman lists three origins: "Whether it's written into your contracts or part of your compliance with PCI or the Executive Order on Improving the Nation's Cybersecurity". It also suggests sharing the SBOM "with relevant stakeholders to build trust". If a customer contract or a security questionnaire asks for one, that is a contract matter: answer what you produce today and check the contract before promising a format or a delivery schedule.
- How fast do dependency vulnerabilities have to be fixed?
- The fetched sources give no SOC 2 deadline. Vulert prints example timelines and says of them: "These timelines are not universal SOC2 rules, but they give auditors something concrete to test." The deadline that counts is the one in your own policy. This page does not repeat Vulert's example numbers; the generated policy has its own, quoted above, and you should change them before approval if you cannot meet them.
Policyseed provides governance policy templates and AI tailoring. It is not legal advice and not a compliance guarantee. Management adopts the policies; the CPA firm performs the SOC 2 examination.