Risk acceptance vs risk exception

A risk acceptance records who agreed to live with a risk. A risk exception records that an asset may depart from a rule, plan or setting. Two hosts missing the 2 Nov 2026 deadline for CVE-2026-80521 show it: one needs an exception only, the other an exception with an acceptance behind it.

  1. Remediate
  2. Mitigate now, fix later
  3. Accept with a dated exceptionThis page is about it

What the primary sources call each one

NIST SP 800-39 defines acceptance as a response to a risk: it "is the appropriate risk response when the identified risk is within the organizational risk tolerance" (Task 3-1). In US federal systems the signer is fixed: "The explicit acceptance of risk is the responsibility of the authorizing official and cannot be delegated" (SP 800-37 Rev. 2, Task R-4).

An exception is a departure from something else. SP 800-40 Rev. 4 uses the word for maintenance plans: "Organizations should closely track and monitor all exceptions to maintenance plans," and "having some exceptions is inevitable" (Section 3.5.5). SP 800-53 Rev. 5 uses "deviations": CM-6 c requires organizations to "Identify, document, and approve any deviations from established configuration settings" (CM-6). Neither text says who approves, or whether residual risk must be accepted.

Uses of the two words in the primary sources; all read 19 Nov 2026. The 800-53 clauses are from the OSCAL catalog of Rev. 5 (source).
TermWhat it governsWho decides in the textSource
Risk acceptanceA residual risk that staysThe authorizing official for US federal systems; the policy elsewhereSP 800-39 Task 3-1; SP 800-37 Rev. 2 Task R-4
Exception to a maintenance planAn asset that cannot follow its patch planNot stated; track and review regularlySP 800-40 Rev. 4, Section 3.5.5
Deviation from a configuration settingA setting that differs from the baseline"approve" is required; the role is not namedSP 800-53 Rev. 5, CM-6 c
Permit by exceptionA deny-all rule with a list of allowed programs or trafficNot a risk decision; a controlSP 800-53 Rev. 5, CM-7(5) and SC-7(5)
Waiver or exception to a federal directiveNot availableNobody: CISA "does not issue waivers or exceptions"BOD 26-04 guidance

SP 800-40 Rev. 4 also pushes against exceptions as a habit. "Instead of denoting certain assets or types of assets as 'exceptions,' there should be maintenance groups for them," and assets with similar long-term exceptions "might need to be moved to a separate maintenance group with its own maintenance plan." A medical device or legacy controller that is renewed every quarter as an exception is telling the team to build a plan for that class.

Which record does a late patch need

Decision table for choosing the record. The sources name no approver for exceptions, so the policy must; the first row assumes it does.
Is the asset departing from a rule or plan?Does a risk remain that someone must own?Record needed
YesNo: the control or the fix closes the gap inside the exception periodException only, approved by whoever owns the plan
YesYes: the gap is open during the exceptionException, plus a signed acceptance with its own expiry
NoYes: a risk is kept and no rule is bentAcceptance only
NoNoNeither; fix it in the normal cycle

The two records have different owners and lifetimes. An exception ends when the plan can be followed or the period lapses. An acceptance ends at its own date or when a reopen trigger fires. One exception can rest on an acceptance, and the exception register links the two rows; the risk register versus exception register guide covers the layout.

Two hosts, one late patch

CVE-2026-80521 is a use-after-free in the Linux kernel's AF_UNIX garbage collector. An attacker must already run code on the host as an unprivileged user and win a narrow race. The 6.1 stable fix shipped in Linux 6.1.189 on 3 Oct 2026 (changelog), and the CVE was not in the CISA catalog in version 2026.11.18 (NVD record, KEV feed). The maintenance plan below is invented: kernel fixes within 30 days of release, so the due date is 2 Nov 2026.

Host B needs an exception only; Host A needs an exception with an acceptance behind it. The CVE facts and fix date are real; the hosts, the plan, the people and the approvals are invented.
Host B, backup applianceHost A, shared build server
Who can run code locallyThree administrators; no unprivileged usersAbout 40 developers with shell accounts
Why the plan is missedReboot needs a window; the next is 14 Nov 2026Release freeze until 15 Dec 2026
Late by12 days43 days
Risk that remainsNone beyond what the three administrators already holdAn unprivileged user could reach the flaw while the patch is waiting
RecordException to the maintenance plan, approved by the plan owner on 30 Oct 2026Exception requested 28 Oct 2026; acceptance signed by the security head on 29 Oct 2026, expiry 15 Dec 2026
Controls namedNone neededDeveloper logins limited to a jump host; unusual process creation by the build user alerted; both checked 29 Oct 2026
Reopen whenWindow moves againThe CVE reaches the catalog; either control fails
Closed14 Nov 2026, after the kernel version checkPlanned 15 Dec 2026

Writing "exception" on Host A's record would have left the signer's name off it. A reviewer reading it would learn which rule was bent but not who accepted the exposure of 40 accounts for 43 days. Writing "acceptance" on Host B's would have sent a signer a decision with no risk in it. The word on the record tells a reader which question was answered. How the end of each period is handled is in risk exception expiry, and the tests behind Host A's signature are in risk acceptance criteria.

Sources

Last reviewed