What to do when a patch would break a critical system

Mitigate now and fix on a dated plan, unless the flaw is exploited and exposed. NIST SP 800-40 Rev. 4 accepts testing and outage windows as reasons to wait, but PHP2Shell (CVE-2026-109618) reached the KEV on 26 Oct 2026 with a 29 Oct due date, and CISA grants no waivers.

  1. Remediate
  2. Mitigate now, fix laterGuidance points here
  3. Accept with a dated exceptionAlso allowed

Signals: Testing shows regression; An outage window is not available yet

The situation and the decision

A vendor patch exists. A test, a pilot or a past incident shows that installing it would disrupt a system the business cannot lose, or no outage window is available yet. The decision is to put an interim control on the exposure now, record the test failure, and set a dated plan for the patch. How long that plan can run depends on whether the flaw is exploited and who can reach the system.

NIST names three different troubles under the word "broken", and each calls for a different record:

Categories from NIST SP 800-40 Rev. 4, Section 3.3, read 19 Nov 2026.
What is wrongResponseWhat the record holds
The patch is flawed and does not correct the vulnerabilityEmergency mitigation; vendor ticket; re-check when the vendor reissuesVendor reply, the re-test result
The patch disrupts other software or systemsEmergency mitigation; fix the conflict or schedule an outage windowRegression evidence, new test plan, next window date
The patch itself is compromisedIncident response, outside this pageIncident number

What the sources allow

Quotes read 19 Nov 2026.
SourceWhat it allows or requiresQuote
NIST SP 800-40 Rev. 4, Section 2.1Waiting for testing and outage windows"The organization may need to wait for a scheduled outage window, perform testing first, update other software that interacts with the software to be patched, or train employees on new features or interfaces."
NIST SP 800-40 Rev. 4, Section 3.3Emergency mitigation when the patch is the problem"Emergency mitigations are sometimes needed because of issues with a patch."
NIST SP 800-40 Rev. 4, Section 3.5.3The mitigation is temporary"Schedules should be set and enforced for both patch deployment and mitigation removal."
NIST SP 800-53, SI-2 bTesting is required"Test software and firmware updates related to flaw remediation for effectiveness and potential side effects before installation;"
CISA, BOD 26-04 implementation guidanceNo waiver for federal directive dates"CISA does not issue waivers or exceptions for actions required in cyber directives."

For high-availability systems the same guidance says agency authorizing officials "should establish change management processes and continuity of operations plans facilitating remediation timelines in BOD 26-04", for example by putting staff time on manual testing of urgent patches. Section 3.5.2 of SP 800-40 Rev. 4 adds that even in an emergency a patch can go first to a small number of canary assets "to confirm that the patch is not corrupted and does not break the software", a stage of minutes to hours. A canary is the cheapest way to find the regression before it reaches the critical system.

A worked case: PHP2Shell, CVE-2026-109618

CVE-2026-109618 is an out-of-memory flaw in PHP that allows unauthenticated remote code execution (CVSS 3.1 base score 9.8). The PHP Group released 8.5.12, 8.4.27, 8.3.36 and 8.2.35 on 21 Oct 2026 at 14:00 UTC. CISA's alert the same day urged immediate patching, and CISA added the CVE and its sibling CVE-2026-109619 to the KEV on 26 Oct 2026, with a due date of 29 Oct 2026 and forensic triage required. PHP 8.1 and older get no fix.

What the patch changes bears on testing. PHP's advisory says it removes longjmp() for out-of-memory errors and replaces it with "a delayed, graceful soft OOM triggered at the next VM safe point, and an immediate hard OOM using _exit() if memory consumption reaches a second, higher threshold". A critical application that runs close to its memory limit is the one to run through staging first.

Suppose an order-processing system on PHP 8.3 fails its staging test on 22 Oct 2026 (an assumed date) because a memory-heavy batch job now ends on the hard limit. The interim controls come from PHP's advisory, and they are incomplete by its own account: setting memory_limit=-1 is "not a complete mitigation" since a failed mmap() can still reach the flaw, and the advisory adds limits on request size and multipart parts.

Release, KEV and tooling times from PHP's advisory, the KEV feed and Fastly's report; the staging row is an assumed example.
DateEventWhat the record shows
21 Oct 2026, 14:00 UTCFixed releases published; exploit tooling already seen at 03:59 UTCPatch exists; clock for the site's own policy starts
22 Oct 2026 (assumed)Staging regression on the memory-heavy batch jobFailed test result attached; interim controls applied and tested
26 Oct 2026KEV listing; due date 29 Oct 2026Federal due date set; the interim control is no longer a plan for an agency
29 Oct 2026Federal due datePatched, or offline; no waiver

Here the usual answer is a bridge of days, not a plan of weeks. CISA's alert, dated 21 Oct 2026 and revised on 23 and 26 Oct, says: "Organizations that cannot update immediately, or that continue to use PHP 8.1 or earlier unsupported versions, should take PHP-based systems offline until they can be patched or updated. A web application firewall may provide some protection until systems are updated or patched." For an agency, the KEV listing made 29 Oct a hard date, and the route when it cannot be met is removal from the network.

What to record, and when it expires

  • The failed test: date, build, what broke and the evidence.
  • The interim control, its test date, and what it leaves open (here, the advisory says the configuration mitigation is not complete).
  • The next outage window or the date the conflict is expected to be fixed, and the approver. The risk acceptance form carries the fields.
  • Expiry: the date of the next outage window. If it passes without the patch, the entry has expired; NIST asks that "All exceptions to maintenance plans should be reviewed regularly to determine if the maintenance plan can be implemented now."

When the usual answer is wrong

  • The flaw is in the KEV and the system is exposed. See patch available and exposed: the due date stands and CISA says no waivers.
  • The interim control is partial and the flaw is exploited. A web application firewall rule is a bridge of days; the WAF rule entry says what it leaves unfixed.
  • Exploitation began before the fix, as with PHP2Shell, where exploit tooling appeared about ten hours before the release. Delay is then measured against live attacks, not a calendar.
  • The patch is flawed rather than disruptive. Mitigation continues, and the record is the vendor ticket, not a test failure.

Sources

Last reviewed