OSSeva Blog
ComplianceHow to Document End-of-Life Software Risk: Risk Acceptance Template for SOC 2 and PCI DSS
The short answer
Document end-of-life software risk as one register entry per unsupported component and version, not as a line in a general IT risk log. Each entry records the component, the version you run, the upstream end-of-life date and where that date came from, how exposed the component is, the known CVEs against it, an owner, a decision, the compensating controls in place, a review date and an exit plan. Where the decision is to keep running the component, attach a time-boxed exception approved by someone with the authority to accept the risk. Then review the entry on a fixed cadence and close it only with evidence that the component was upgraded, replaced, retired or put under a documented source of security fixes.
That structure maps onto what the main frameworks ask for. NIST SP 800-53 SA-22 wants replacement or an alternative source of support, and assessors examine "documented approvals, including justification" for continued use. PCI DSS 12.3.4 wants end-of-life plans documented and a remediation plan approved by senior management. This article is general information, not legal or audit advice. Your auditor or assessor makes the final call on whether the evidence is enough.
In short:
- Keep one register entry per unsupported component and version, each with the end-of-support date and where it came from.
- Record a decision for every entry: upgrade, replace, retire, or keep with a named source of security updates.
- Where the component stays in service, attach a time-boxed exception with an owner, an approver, tested controls and an exit plan.
- Review each entry at least yearly, closer to quarterly for end-of-life software, and close it only with evidence.
Running open source in production? OSSeva patches and supports it, end-of-life versions included.
What is end-of-life software?
End-of-life software is software whose developer, vendor or open source project has stopped supporting a version, so it no longer receives security updates or bug fixes. The date support stops is its end-of-support or end-of-life date. NIST SP 800-53 calls these unsupported system components.
Running Ruby or Rails past end of life? Ruby end-of-life dates and Ruby on Rails end-of-life dates list each version with the date it was checked.
The risk is not that the software stops working. NIST's guidance for control SA-22 says support includes software patches, firmware updates, replacement parts and maintenance contracts, and that once vendors stop providing critical patches or product updates, adversaries get an opportunity to exploit weaknesses in the installed components. A flaw disclosed after the end-of-support date stays open on that version unless someone else supplies the fix, so the component carries more known exposure with each new CVE. That widens its attack surface in practice (NIST's glossary defines the term as the set of points where an attacker can try to enter, affect or extract data from a system or component). It is why an EOL software component needs its own register entry and not just a line in a general IT risk log.
End-of-life risk acceptance template
Copy the form below into your ticketing or GRC tool, one per register entry. It is built so that the same record can answer a SOC 2 request for management's risk decision and a PCI DSS request for the technology review and its remediation plan. No template guarantees an outcome: your auditor or Qualified Security Assessor decides whether the evidence is enough.
END-OF-LIFE SOFTWARE RISK ACCEPTANCE (EXCEPTION)
Exception ID: EX-____ Raised on: ____ Register entry: ____
1. ASSET
Component and version line: ____
Exact version in production (build, SBOM or command output): ____
Environment, hosts, clusters or images: ____
In scope for: [ ] SOC 2 [ ] PCI DSS [ ] other: ____
2. END-OF-LIFE STATUS
Upstream end-of-life date: ____
Source URL and date checked: ____
Newest release on this line: ____
3. EXPOSURE
Reachable from (internet / internal / named segment): ____
Data handled and its classification: ____
Known CVEs for this version, their severity, and whether upstream
has a fix for this line (yes / no): ____
4. CONSTRAINT
Why the component cannot be upgraded or replaced yet: ____
5. RISK STATEMENT
What could go wrong, and the risk that remains after controls: ____
6. SOURCE OF SECURITY FIXES
In-house / external provider / none: ____ Contract or procedure: ____
7. COMPENSATING CONTROLS (one row per control)
Control | Risk it addresses | How it was validated | Last test (date, result) | Control owner
8. OWNER AND APPROVAL
Owner (a named person): ____
Approver (name, role) and the policy that gives that role the authority: ____
Approval date: ____
9. TIME BOX
Start: ____ Expiry: ____ Renewals allowed by policy: ____
Scheduled reviews: ____ Events that force an early review: ____
Conditions that void the exception: ____
10. EXIT PLAN
Target version or replacement: ____ Target date: ____ Change ticket: ____
11. EVIDENCE ATTACHED
[ ] Inventory or SBOM extract [ ] Scan results [ ] Control test results
[ ] Change records [ ] Fix-source contract or procedure
Sections 1 to 3 and 6 to 10 map to the register fields described below, and the filled-in ZooKeeper 3.6 example shows what a completed entry reads like. Section 7 follows the six-part layout of the PCI DSS compensating controls worksheet, which is why it asks how each control was validated and who maintains it.
What do SOC 2 and PCI DSS assessors look for in an end-of-life risk acceptance?
Assessors want to see that the component is inventoried, that its end-of-support date was checked, that an accountable person approved a plan to remove the risk, that patches or compensating controls cover the gap, and that changes are controlled. The table maps each of these to SOC 2 and PCI DSS.
| What the assessor wants to see | SOC 2 (Trust Services Criteria) | PCI DSS v4.0.1 |
|---|---|---|
| The component is known and tracked | Evidence for CC7.1, the criterion on detecting vulnerabilities: how you learn about new CVEs for this component. | Requirement 6.3.2 (inventory of bespoke and custom software and its third-party components, required since 31 March 2025) and 12.5.1 (inventory of in-scope system components). |
| End-of-life status is reviewed and dated | Management's risk assessment, in the CC3 series of criteria, with the risk and its decision on record. | Requirement 12.3.4: technologies reviewed at least every 12 months, including whether they still receive vendor fixes, whether they still support your PCI DSS compliance, and any vendor end-of-life announcements. |
| A plan to remove the risk, approved by someone accountable | An approver whose authority your own policy defines. The auditor tests you against your stated policy. | Requirement 12.3.4 asks for a documented plan, approved by senior management, to remediate outdated technologies, including those with announced end-of-life plans. |
| Patches are applied on time, or there is a reason they cannot be | The patching timeline in your own control description, and evidence it was followed or an exception was approved. | Requirement 6.3.3: applicable patches installed within a defined time, with critical patches due within one month of release. On an ended line nothing is released upstream, so a supplier build or a compensating control has to answer it. |
| Compensating controls exist and were tested | Control descriptions, tests and results that cover the period under audit. | Appendix B compensating controls worksheet: the constraint, the controls, the objective, the added risk, how the controls were validated and how they are maintained. PCI SSC says only entities with a legitimate, documented technical or business constraint can use compensating controls. |
| Changes are controlled | CC8.1, the criterion on change management: change records for patches, upgrades and the final removal. | Requirement 6.5.1: change procedures with a reason, security impact, approval by authorised parties and testing. |
AICPA publishes the Trust Services Criteria behind a free-account download, so this article describes CC3, CC7.1 and CC8.1 in our own words rather than quoting them; read the criteria text with your auditor. The PCI DSS wording above was read in PCI SSC's SAQ D for PCI DSS v4.0 (April 2022). The standard itself sits behind a PCI SSC download agreement, and our PCI DSS end-of-life software guide covers where the v4.0.1 wording of 6.3.3 differs.
Why a risk acceptance fails sampling
- No expiry, or an expired one. An exception with no end date is a permanent waiver. Treat an expired exception as an open finding.
- No named owner or approver. A team alias cannot be interviewed.
- Controls described but never tested. Each control needs a test date and a result.
- The exception stands in for a fix. It records the decision to keep running the component. It does not make the vulnerability go away, so section 6 has to name where fixes will come from, and section 10 has to end the need for the exception.
- The scanner finding was closed with a comment. Link every scanner finding to the exception or to the change that fixed it.
For the policy this form belongs to, see our end-of-life software policy template. Framework-by-framework detail is in our guides for PCI DSS, SOC 2, DORA and NIS2 and FedRAMP and CMMC, and the compliance evidence hub lists what OSSeva supplies for a patched build. Where a scanner keeps flagging a version that carries backported fixes, how audit teams treat patched builds explains why.
What fields should an end-of-life risk register entry have?
Keep one row per component and version line in each environment where it runs. A ZooKeeper 3.6 ensemble in production and another in a payment environment are two rows, because their exposure, owners and controls differ.
| Field | What to record | Why an auditor asks for it |
|---|---|---|
| Component | Product name and where it runs: cluster, service, image or host group. | Ties the entry to the inventory (NIST CM-8, PCI DSS 12.5.1). |
| Version | The exact version in production, taken from the build or SBOM, not from a design document. | End-of-life status applies to a version line, and scanners match on version. |
| Upstream EOL date and source | The date security fixes stopped, plus a link to the project's announcement or release history. | Shows the date was checked, not guessed. PCI DSS 12.3.4 asks you to document vendor end-of-life plans. |
| Exposure | Internet-facing or internal, which networks can reach it, what data it handles, which frameworks put it in scope. | Exposure drives likelihood in the risk assessment (NIST RA-3). |
| Known CVEs | Published CVEs affecting this version, their severity, whether a fix exists for this line, and any VEX status. | Shows the risk is assessed against real vulnerabilities, not just age. |
| Owner | A named person, not a team alias, accountable for the decision and the exit plan. | Assessors interview the people responsible for component replacement under SA-22. |
| Decision | Upgrade, replace, retire, or keep with a named source of fixes, plus the exception reference if kept. | SA-22 accepts replacement or an alternative source of support; this field says which. |
| Compensating controls | Each control, the finding it addresses, and the date and result of its last test. | A control nobody has tested is a description, not evidence. |
| Review date | When the entry is next reassessed, and the triggers that force an earlier review. | Shows the risk is managed over time, not accepted once. |
| Exit plan | Target version or replacement, target date, dependencies and the change ticket. | PCI DSS 12.3.4 asks for a plan to remediate outdated technologies; NIST CA-5 asks for a plan of action and milestones. |
The register sits next to your SBOM rather than inside it. NTIA's minimum SBOM elements have no field for support status, which is why our post on SBOMs and end-of-life components argues that an inventory finds the problem without fixing it. Our comparison of end-of-life tracking tools covers how to populate the date and source columns automatically.
Accept, mitigate, upgrade or buy extended support: which decision fits?
Choose the option that removes the vulnerability when you can, and use a time-boxed exception only for the gap before it lands. Upgrading, replacing or retiring ends the risk. Extended support and compensating controls reduce it. Accepting the risk with no other step leaves it exactly where it was.
| Decision | What it means | Removes the vulnerability? | What the register entry needs | When it fits |
|---|---|---|---|---|
| Upgrade or replace | Move to a supported version or product. | Yes, once complete. | Target version, date, change ticket; close the entry with a scan after the change. | The default whenever the dependencies allow it. |
| Retire | Decommission the component and its data. | Yes. | Decommissioning record; images and disaster-recovery copies removed. | Nothing depends on it any more. |
| Keep with extended support | Stay on the version and take security updates from an in-house team or an external provider. | Covers the CVEs the provider fixes; the version is still past upstream end of support. | Named source of fixes and its contract or procedure (SA-22 part b), exception, exit plan, review dates. | The upgrade is blocked or will take longer than the exposure allows. |
| Mitigate with compensating controls | Reduce who can reach the component and what an attacker can do with it. | No. It lowers likelihood or impact. | Each control, the finding it addresses, test date and result; exception and exit plan. | A short bridge, or a layer on top of extended support. |
| Accept the risk | Management signs off on running the component as it is. | No. | Risk statement, named approver with authority, expiry date, conditions that void it. | Low exposure and low data value, with an expiry. Assessors rarely accept it as the only measure on an exposed component. |
Most real entries combine two rows: an exception that bridges the gap, plus a source of fixes or a set of tested controls while the upgrade is scheduled.
A filled-in example: ZooKeeper 3.6
Apache ZooKeeper 3.6 reached end of life on 30 December 2022, with 3.6.4 as its final release, as recorded on our ZooKeeper 3.6 end-of-life page. The deployment details below are illustrative.
| Field | Example entry |
|---|---|
| Component | Apache ZooKeeper ensemble zk-core, three members, coordinating two internal Kafka clusters and a service-discovery layer |
| Version | 3.6.4, confirmed with the srvr command on each member |
| Upstream EOL date and source | 30 December 2022. Apache ZooKeeper release history, cross-checked against endoflife.date |
| Exposure | Internal only. Reachable from the application subnet. Holds service metadata, no cardholder or patient data. In scope for SOC 2. |
| Known CVEs | CVE-2024-23944 (rated critical by Apache): persistent watchers can expose child znode paths a client should not see. The advisory lists 3.6.0 through 3.7.2 as affected and fixes it only in 3.8.4 and 3.9.2, so there is no upstream fix for 3.6. |
| Owner | Platform engineering lead (named) |
| Decision | Keep on 3.6 with an external source of backported fixes until the upgrade completes. Exception EX-0142. |
| Compensating controls | Client connections limited to named hosts and SASL principals; ACLs reviewed so no client can read parents of sensitive subtrees; four-letter-word commands restricted. Last tested on the date of the exception, results attached. |
| Review date | Quarterly, and on any new CVE that names 3.6 or an earlier line |
| Exit plan | Rolling upgrade to 3.9 after every client is inventoried; change ticket linked; target date set in the exception |
Two details make this entry hold up. The CVE field says plainly that upstream has no fix for this line, so nobody reads a clean scan as a clean bill of health. And the decision names where fixes will come from, which is the question an SA-22 assessor asks first. Our SA-22 guide lists the evidence for each route.
What should a time-boxed exception contain?
An exception, sometimes called a risk acceptance, is the signed decision to keep running something that policy says should be upgraded. It is separate from the register entry because it has a different lifecycle: it is approved once, expires, and is either closed or renewed with fresh justification. A useful exception contains:
- Scope. The register entry it covers, the component, version and environments. One exception per entry keeps renewals honest.
- The constraint. Why the component cannot be upgraded yet: client dependencies, vendor certification of a dependent product, a change freeze. PCI DSS's compensating controls worksheet asks for the legitimate technical or business constraint first.
- The risk statement. What could go wrong, using the exposure and CVE fields, and what risk remains after controls. The PCI worksheet calls this the additional risk posed by not meeting the original requirement.
- The source of fixes. In-house team or external provider, and the contract or procedure that defines it. This is SA-22 part b.
- Compensating controls and their tests. Each control, how it was validated, and how it will be maintained.
- Approver. Someone with authority to accept the risk at its level, named in your policy. SP 800-53A looks for documented approvals with justification, and PCI DSS 12.3.4 asks for senior-management approval of the remediation plan.
- Start date, expiry date and maximum duration. Expiry is what makes it time-boxed. Set a maximum duration and a limit on renewals in your policy, so that a long-running exception has to be argued again on its merits.
- Conditions that void it early. For example, a new CVE on the component listed in CISA's Known Exploited Vulnerabilities catalog, a change that exposes the component to the internet, or a lapse in the support contract.
- Link to the exit plan. The ticket or plan of action and milestones (NIST CA-5) that ends the need for the exception.
An expired exception should be treated as an open finding, not quietly extended. Our compliance versus security post explains why: an accepted exception closes the audit finding but not the risk.
Compensating controls auditors usually accept, and their limits
Auditors generally accept compensating controls for an end-of-life component when they reduce a specific risk, have been tested, and come with a plan to stop needing them. They are much less comfortable with controls offered as a permanent substitute for support. The common ones:
| Control | What it reduces | Limits |
|---|---|---|
| Network isolation | Who can reach the vulnerable service. The SA-22 discussion names prohibiting connection to public or uncontrolled networks as a mitigation. | Holds only while the network design stays the same. Isolation reduces risk but is not one of SA-22's two ways of complying. Cyber Essentials accepts unsupported software only if it is removed or placed in a sub-set that prevents all traffic to or from the internet. |
| Web application firewall | Known attack patterns against HTTP interfaces. | Covers only traffic that passes through it and only signatures it knows. No help for non-HTTP protocols such as ZooKeeper's client port, or for attacks from inside the trusted zone. |
| Virtual patching (IPS or WAF rules for a specific CVE) | Exploitation of one named vulnerability, often within days of disclosure. | Needs a rule per CVE, and the vulnerable code is still present. Evasion and protocol variants are a standing concern, so auditors expect a fix to follow. |
| Monitoring and detection | Time to notice an attack in progress. | Detects; does not prevent. Acceptable as a layer, rarely as the only control for an exposed component. |
| Backported security patches | The vulnerability itself, on the version you already run. | Strictly a fix rather than a compensating control, and it covers only the CVEs patched. It satisfies SA-22 part b when a contract or in-house procedure defines it, and it still needs change records and a register entry until the upgrade. |
Whichever controls you use, record for each one the finding it addresses, the test, the result and the date. A firewall rule that was correct before the last network change is the usual reason a compensating control fails sampling.
How often should you review end-of-life software risk?
At least annually for every entry, because that is the floor in the frameworks that set one. PCI DSS 12.3.4 requires a review of hardware and software technologies at least once every 12 months. DORA Article 8(7) requires financial entities, other than microenterprises, to run a specific ICT risk assessment on all legacy ICT systems "on a regular basis, and at least yearly", and before and after connecting new technologies, applications or systems. DORA's definition of a legacy ICT system in Article 3(3) includes one that has reached end of life or is no longer supported by its supplier or an ICT third-party service provider.
Annual is the floor, not the target. A quarterly review suits most end-of-life entries, with an immediate review when one of these happens:
- A new CVE names the component's version line, or upstream publishes an advisory without listing older lines.
- A CVE affecting the component is added to the Known Exploited Vulnerabilities catalog.
- The component's exposure changes: a new network path, a new data type, a new framework in scope.
- The exception is within a month of expiry.
- The source of fixes changes, including a contract renewal or a team reorganisation.
How do you close an end-of-life risk item?
Close a register entry only when the risk it records no longer exists: the component was upgraded, replaced or retired, and the evidence is attached. Moving it onto extended support changes the decision but does not close the entry. Attach the evidence that fits the route:
- Upgraded. The change record, the SBOM or inventory showing the supported version, and a scan after the change. Then close the exception.
- Replaced. The same evidence for the replacement product, plus confirmation that the old component no longer runs anywhere, including disaster-recovery copies and base images.
- Retired. The decommissioning record and evidence that the hosts, images and data were removed or archived.
Moving a component onto extended support does not close the entry. It changes the decision to "kept, with an external source of fixes" and the review continues, because new CVEs will keep arriving against an old line. What changes is that each new CVE now has somewhere to go. Our audit evidence checklist shows how the closed entry, the change records and the approvals fit into one package for sampling.
Where OSSeva fits
OSSeva is the external source of fixes that the decision field can name for community end-of-life runtimes. OSSeva Patch delivers drop-in builds with backported security patches quarterly and out of cycle for CVSS 9.0 and above, signed with GPG and attested with Sigstore Cosign. For ZooKeeper that includes patched 3.6.x builds. OSSeva Operate commits to patching Critical CVEs (CVSS 9.0 or higher) within 48 hours and High within 7 days. OSSeva Assure adds architecture review and compliance documentation. Our end-of-life dates and EOL tracker give you the date and source for the register.
Frequently asked questions
What is the best way to document software end of life risk?
Keep one risk register entry per unsupported component and version, with the upstream end-of-life date and its source, exposure, known CVEs, an owner, a decision, compensating controls, a review date and an exit plan. Link a signed, time-boxed exception to any entry where the component stays in service.
How do you manage end of life software compliance?
Track support status alongside your inventory, decide for each end-of-life component whether to upgrade, replace, retire or keep it with a documented source of fixes, and review the decisions on a fixed cadence. Our end-of-life software policy template sets out the rules in a form you can adopt.
How do you prepare for an end of life software audit?
Make sure every end-of-life component in scope has a register entry, a current approval with justification, evidence of where its fixes come from, tested compensating controls and change records for fixes applied in the period. Auditors sample, so assemble the evidence per component rather than per framework.
How do you reduce risk from obsolete software?
Contain it first with isolation and access restrictions, get a source of security fixes for the version you run, and schedule the upgrade or retirement. Controls lower the risk quickly; only a fix or removal takes the vulnerability away.
Is a risk acceptance enough on its own?
Rarely. Frameworks allow documented risk acceptance, but auditors expect it to be time-boxed, approved at the right level, backed by tested controls and linked to a plan that ends it.
Sources
Checked on 28 October 2026, except where a line says 8 November 2026.
- PCI SSC, Self-Assessment Questionnaire D for Merchants, PCI DSS v4.0 (April 2022), requirements 6.3.2, 6.3.3, 6.5.1, 12.3.4 and 12.5.1 and Appendix B (compensating controls worksheet): listings.pcisecuritystandards.org. The v4.0.1 standard (PDF) could not be fetched, so wording differences are noted in our PCI DSS guide.
- AICPA, 2017 Trust Services Criteria with revised points of focus (2022): aicpa-cima.com. The download page was reached; the criteria text was not, so CC3, CC7.1 and CC8.1 are described, not quoted.
- NIST SP 800-53 Rev. 5, control SA-22 (Unsupported System Components) and its guidance, from the OSCAL catalog: usnistgov/oscal-content, and the NIST CSRC glossary entry for attack surface. Both checked on 8 November 2026.
Tags
Related articles
Open Source Compliance: 9 Audit-Ready Evidence Checks
October 6, 2026ComplianceOpen Source Compliance vs Security: What Teams Need
October 6, 2026ComplianceTools for Tracking End-of-Life Software Across an Organization
October 6, 2026ComplianceNode.js 18 and 20 in PCI DSS and SOC 2 Audits: What Assessors Ask
November 17, 2026ComplianceTomcat 9 after end of life in a PCI DSS 4.0 audit: what the assessor asks for
November 17, 2026ComplianceEnd-of-Life Software Policy Template for Open Source Infrastructure
October 7, 2026Get a quote
Ready to get your open source under control?
Tell us what you run. An OSSeva engineer replies within 1 business day with CVE coverage and a price for your stack.
Prefer to talk or read more first?