Remove internet exposure as a vulnerability mitigation

Removing a product from the internet is a mitigation CISA names in BOD 26-04, but it shortens the clock only in some rows of Table 1. For Atlassian's CVE-2026-21589, added to KEV on 15 Oct 2026, it moves 3 days with forensic triage to 14 days. For FortiMail's CVE-2026-104286 it changes nothing.

What this control does toward each outcome

  1. RemediateLeaves the flaw: Does not fix the flaw. Internal attackers and anything that still reaches it remain. It can be undone by a later network change.
  2. Mitigate now, fix laterLowers exposure: Takes the component off public networks (pull the listener, put it behind VPN or a private link). BOD 26-04 treats this as changing the exposure rating, which moves the required timeline.Guidance points here
  3. Accept with a dated exceptionIf the fix date slips: accept the residual risk in a signed, dated, reviewed exception.Also allowed

Decision Table row: A control already sits in front of the vulnerable component (WAF rule, segmentation, access restriction). Group: Who can reach it.

Evidence to keep: External connectivity check result with date; Before and after exposure inventory entry; Change ticket.

What CISA says removal does

BOD 26-04 defines the input this way: "Publicly Exposed is defined as Yes if the asset in scope of this directive is accessible to unauthenticated or untrusted entities via public networks, such as the internet, regardless of its physical or logical location; the value is No otherwise." Its notes to Table 1 then say: "one valid mitigation is to remove the system from the internet; that action changes the value of "Publicly Exposed" from Yes to No and will shift the required timeline for further action, such as remediation, back." (CISA BOD 26-04, Appendix A, read 19 Nov 2026.)

Two limits sit in the same text. If any method used to check exposure finds the asset exposed, the value is Yes. And the implementation guidance leaves the final call to the agency, asset by asset (BOD 26-04 implementation guidance, read 19 Nov 2026). A product behind a VPN that only authenticated staff can use is for the agency to rate. A product that still answers an internet address is exposed.

CISA's industrial control advisories give the same advice as a recommended practice: "Minimize network exposure for all control system devices and/or systems, ensuring they are not accessible from the internet", and when remote access is required, use a VPN while "recognizing VPNs may have vulnerabilities and should be updated to the most current version available". The same advisory adds: "CISA reminds organizations to perform proper impact analysis and risk assessment prior to deploying defensive measures." (CISA ICSA-26-321-11, a republication of Siemens SSA-724261, read 19 Nov 2026.)

Which rows of Table 1 move

Table 1 has 16 rows built from four inputs: Publicly Exposed, In the KEV, Automatable by Adversary, and Technical Impact. Changing only the first input pairs each exposed row with the unexposed row that has the same other inputs. Read from CISA's published table on 19 Nov 2026:

BOD 26-04 Table 1, Appendix A, read 19 Nov 2026. Days are calendar days. For a KEV entry whose automatable value is yes and whose impact is total, removal buys no time.
In the KEV / Automatable / ImpactExposed (row, days)Not exposed (row, days)Effect of removal
Yes / Yes / Total1: 3 days & forensic triage9: 3 days & forensic triageNone
Yes / Yes / Partial2: 3 days10: 14 days3 days become 14
Yes / No / Total3: 3 days & forensic triage11: 14 days3 days & forensic triage become 14 days
Yes / No / Partial4: 14 days12: 14 daysNone
No / Yes / Total5: 3 days13: 60 days3 days become 60
No / Yes / Partial6: 14 days14: 60 days14 days become 60
No / No / Total7: 14 days15: fix on system upgrade14 days become the next scheduled upgrade
No / No / Partial8: 60 days16: fix on system upgrade60 days become the next scheduled upgrade

Four KEV entries worked through Table 1

The automatable and impact values below are CISA's own, published in the NVD record (CISA Coordinator, SSVC version 2.0.3). The KEV due dates and forensic triage flags are from the KEV feed, catalog 2026.11.18. Each exposed row matches the KEV due date and flag, which is how the row was identified.

CISA KEV feed 2026.11.18, NVD records for the four CVEs (CISA ADP SSVC data), and BOD 26-04 Table 1, all read 19 Nov 2026.
CVE and productKEV added, dueAutomatable, impactExposed rowUnexposed rowClock change
CVE-2026-21589, Atlassian Data Center products15 Oct 2026, 18 Oct 2026; triage yesNo, total3: 3 days & forensic triage11: 14 daysDue date moves from 18 Oct to 29 Oct 2026
CVE-2026-104286, Fortinet FortiMail1 Oct 2026, 4 Oct 2026; triage yesYes, total1: 3 days & forensic triage9: 3 days & forensic triageNone
CVE-2026-76460, Cisco Identity Services Engine16 Sep 2026, 19 Sep 2026; triage yesYes, total1: 3 days & forensic triage9: 3 days & forensic triageNone
CVE-2026-19868, SonicWall SonicOS SSLVPN6 Nov 2026, 20 Nov 2026; triage noNo, partial4: 14 days12: 14 daysNone

Only the Atlassian entry gains days. Exposure removal on the other three is still a sound way to cut the chance of compromise while the upgrade is booked, but under the directive it does not change the due date, and the forensic triage requirement on FortiMail and ISE stays.

Worked case: Atlassian CVE-2026-21589

Atlassian published its advisory on 5 Oct 2026 and wrote: "Remove your instance from the internet until you can patch or apply mitigations, if possible. Instances accessible to the public internet, including those with user authentication, should be restricted from external network access until you can take action." CISA added the CVE to KEV on 15 Oct. A federal agency that had pulled its Confluence Data Center off the internet by 15 Oct, and could show it, would move from row 3 to row 11: the due date from 18 Oct to 29 Oct 2026 (14 days from 15 Oct), and the forensic triage text no longer attached to the row.

Three facts keep that from being a bargain. The instance was reachable from 5 Oct, ten days before KEV, and Atlassian says it cannot confirm whether instances were affected and tells you to check all of them for signs of compromise, so the triage still happens. The fix is still due. And the cost is that external users and integrations lose access for as long as the instance is off the internet. A team that still needs those users can use Atlassian's WAF rule instead, which does not change the Table 1 row.

FortiMail shows the other side. Fortinet's advisory FG-IR-26-175 lists, besides turning the IBE service off, "Disable access to the FortiMail webmail interface from the internet or limit the access only from trusted private network". The cost is that external users, including IBE recipients, lose webmail and the secure portal. Under Table 1 the due date stays 4 Oct 2026. Fortinet fixed its hosted FortiMail Cloud on 5 Oct 2026, so only self-managed appliances carry this finding.

How to verify the exposure is gone

  1. List every way the product was reached: public DNS names, NAT and port-forward rules, load balancer listeners, cloud security groups, CDN origins and VPN gateways. Remove or restrict each and keep the before and after export, dated.
  2. From a network you do not control, such as a mobile hotspot, request the product's own address and hostname and record that nothing answers. This is a reachability check of your own address space, not a test of the flaw.
  3. Check DNS and certificate records for stale names that still resolve to the product.
  4. Update the asset's exposure tag. BOD 26-04 Phase III has agencies tag assets "that can be reached from outside the agency network and using a routable IP address" and report every seven days when reporting is not automated; the tag change is what moves the timeline.
  5. Re-check after every network change and put the check on the change template.

What to record

Exposure removal entry

  • Finding: CVE, product, version, KEV added and due dates
  • Table 1 inputs: automatable, impact; exposed row and unexposed row, with days for each
  • What was removed: DNS names, NAT rules, listeners, security groups, with before and after exports and dates
  • Replacement access path (VPN, private link) and its owner and patch state
  • Outside-network check: date, vantage point, result
  • Exposure tag change date
  • Dependents checked: partners, integrations, remote support, monitoring
  • Forensic triage ticket where the KEV entry requires it
  • Fix date and the date exposure may return

Exposure removal stops being enough when the business needs the product reachable again, when an internal path is the likely route in, or when the flaw is exploited before login by anyone already on the internal network. Then use network segmentation or access restriction, and see the decision row for a KEV finding with no patch yet. Checks on any mitigation are in how to verify a mitigation works.

Sources

Last reviewed