Not internet-reachable, not on the KEV, not automatable

Accept with a dated exception. CISA BOD 26-04 Table 1 gives rows 15 and 16 the timeline "Fix on system upgrade": asset not publicly exposed, CVE not in the KEV, exploitation not automatable. Guidance says agencies need not prioritize them, but a flip in any answer moves the row to 14 or 60 days.

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

Signals: Publicly exposed: no; In the KEV: no; Automatable: no

The situation and the decision

Three answers are all no: the asset is not publicly exposed, the CVE has no entry in CISA's Known Exploited Vulnerabilities (KEV) catalog, and an adversary cannot automate every step of exploitation. The response guidance points to is to accept the finding with a signed, dated exception and let the fix ride the next scheduled major upgrade or rebuild. Mitigating now and fixing later is also allowed.

The decision holds only while the three answers hold. Table 1 of BOD 26-04 shows what one flipped answer does to each of the two "Fix on system upgrade" rows (technical impact is total control in row 15, partial control in row 16):

BOD 26-04 Table 1, read from CISA's published table image on 19 Nov 2026. Timelines are calendar days.
RowBaselineIf it becomes publicly exposedIf it is added to the KEVIf exploitation becomes automatable
15 (total control)Fix on system upgradeRow 7: 14 daysRow 11: 14 daysRow 13: 60 days
16 (partial control)Fix on system upgradeRow 8: 60 daysRow 12: 14 daysRow 14: 60 days

How each answer is decided

Publicly exposed. The directive sets the value to yes if the asset is "accessible to unauthenticated or untrusted entities via public networks, such as the internet, regardless of its physical or logical location". If any one of several overlapping methods finds it exposed, the value is yes, so one scanner saying no does not outweigh another saying yes. Where an agency has not supplied exposure data, CISA's implementation guidance treats an asset as public if it uses a publicly routable IP address or has the asset type "network", and as internal only after those tests fail.

In the KEV. The CVE ID has an entry in the catalog. Catalog version 2026.11.18, read on 19 Nov 2026, holds 1,786 entries.

Automatable and technical impact. Both come from CISA's Vulnrichment data for the CVE. The guidance adds a default: when that data is missing and the CVE is not in the KEV, CISA treats automatable as "no" and technical impact as "total" until CISA or a third party supplies it. An unenriched CVE on an unexposed asset therefore lands on row 15.

Internal is not the same as unreachable. FedRAMP's 2026 evaluation rules focus on "internet-reachable (rather than internet-accessible)" flaws because a service behind a load balancer can still receive a payload that the vulnerable code processes, and they give SQL injection as a classic example on systems "not typically internet-accessible". Once a filter that deflects the triggering payload is in place, the same text says the flaw "should no longer be considered an internet-reachable vulnerability".

The rule it follows

BOD 26-04, Appendix A: "Fix on system upgrade means that, unless conditions change as described in item (e) above, the vulnerability should be remediated the next time the vulnerable asset receives a scheduled major upgrade or rebuild." Item (e) is the reset clause: removing a system from the internet moves it to a slower row, but a CVE added to the KEV "will shorten according to Table 1".

The implementation guidance puts it plainly: "agencies do not need to prioritize these vulnerabilities and can wait for the next system upgrade". It also says the timelines "represent maximum duration until a remediation is applied" and that an agency with lower tolerance "may set shorter timelines".

For anyone outside the federal rule, the test is NIST SP 800-39: "Risk acceptance is the appropriate risk response when the identified risk is within the organizational risk tolerance." The guide on risk acceptance versus risk tolerance covers who sets that limit.

A worked case: CVE-2026-16723 in fastjson 1.x

CVE-2026-16723 is a remote code execution flaw (CWE-502, CVSS 3.1 base score 9.0 from the CNA) in fastjson 1.2.68 through 1.2.83, a Java JSON library. Alibaba's advisory is dated 21 Jul 2026. NVD published the record on 23 Jul 2026, and the same day CISA's coordinator entry on it read exploitation none, automatable no, technical impact total. Alibaba released fastjson 1.2.84 with the fix on 29 Jul 2026 and archived the 1.x repository the same day. The CVE is not in KEV catalog 2026.11.18.

Take a batch service that parses JSON from an internal queue, runs on a private address, and is not tagged as a network device. The advisory's trigger is a Spring Boot executable fat-jar that passes attacker-supplied JSON to the parser. Suppose the agency's tooling first lists the CVE on that service on 29 Jul 2026 (an assumed date for the example). The answers are exposure no, KEV no, automatable no, technical impact total: row 15, fix on system upgrade.

Dates from NVD, the CISA entry on the CVE record, Alibaba's release notes and the KEV feed, read 19 Nov 2026. The last two rows are computed from the assumed 29 Jul start.
DateEventEffect on the row
21 Jul 2026Alibaba publishes the advisoryNone yet
23 Jul 2026NVD record; CISA entry: exploitation none, automatable no, technical impact totalRow 15 for an unexposed asset
29 Jul 2026fastjson 1.2.84 released; 1.x repository archivedFix exists; no further 1.x fixes will come
19 Nov 2026KEV catalog 2026.11.18 checked: no entryRow 15 stands
If exposed on 29 Jul 2026Row 7 appliesRemediate by 12 Aug 2026
If automatable on 29 Jul 2026Row 13 appliesRemediate by 27 Sep 2026

The exception names the next scheduled rebuild of the service as the fix date. Because fastjson 1.x is archived, that rebuild has to be a move to 1.2.84 or to fastjson2 (2.0.63 or later, which the advisory says is not affected by this CVE), so the plan names which one.

What to record

  • The three answers, the method behind each, and the date of each check (the exposure check, the KEV feed version, the date the Vulnrichment values were read).
  • The upgrade or rebuild the fix waits for, with a calendar date even if approximate, and the version it moves to.
  • The triggers that reopen the entry: an exposure change, a KEV listing, a change in the Vulnrichment values, or a new proof of exploitation.
  • The approver. On a federal system under BOD 26-04 the timeline is the rule and nobody signs a longer one; the authorizing official (the role NIST SP 800-37 gives the explicit acceptance of risk) signs only an agency's own written exception. Elsewhere, the risk owner whose stated tolerance covers the finding signs, with a title and a date.
  • A review date. BOD 26-04 sets none for an agency; it asks CISA to reassess its own timelines once per fiscal year, a cadence that is easy to copy. This site's rule: use the sooner of the planned upgrade and a yearly exposure re-check. The risk acceptance form holds the fields and the exception register tracks the entry.

When accepting is the wrong answer

  • The fix is cheap. For CVE-2026-16723 the fix is a dependency bump to 1.2.84 on a service already scheduled for a release, so a dated exception costs more paperwork than the change. BOD 26-04 permits waiting; it does not ask for it.
  • The service is reachable through something in front of it. If the batch queue is fed by an internet-facing form, the parser still handles attacker-supplied data and the exposure answer should be yes under the FedRAMP reading above.
  • One of the three answers rests on a single source. An asset tagged internal from a spreadsheet but seen on a public address by a scanner is exposed.
  • Exploitation is automatable and a patch exists. Automation, not exposure, is the test here. Rows 13 and 14 (not exposed, not in the KEV, automatable) give 60 days, so the 60-day internal row applies and the exception route is closed.
  • The finding may be absent rather than unreachable; see false positive handling.

Sources

Last reviewed