In this Field Note
A router can forward traffic beautifully while somebody else changes its configuration. The uptime graph is not qualified to give it a clean bill of health.
The September 2026 RouterOS disclosures require prompt attention. CERT Polska reports active exploitation of internet-reachable SSH services through a chain of authentication and privilege flaws. For a WISP, the job is to reduce exposure, install an appropriate fixed release, and establish whether the configuration is still yours.
Technical review: September 20, 2026. This is dated operational guidance, not a promise that a release remains the newest or best choice indefinitely. Check the linked vendor and researcher notices before acting.
Three questions, not one green checkbox
| Question | Evidence to collect | What it does not prove |
|---|---|---|
| Is the software fixed? | Running version, branch, and relevant affected ranges. | That the router was never compromised before the update. |
| Who can reach the vulnerable paths? | Effective service settings, firewall rules, and authorized reachability tests. | That a private address is inaccessible to subscribers or contractors. |
| Has someone changed the device? | Configuration history, accounts, logs, and suspicious state. | That missing logs or an absent warning establish a clean history. |
An exposed, unpatched router needs urgent containment and patching. A patched router with unexplained privileged changes needs incident handling. Those jobs overlap, but they are not interchangeable.
Choose by branch, not by the largest-looking number
MikroTik’s initial bulletin listed fixes including 7.24.2, 7.23.4, and 6.49.21. The later per-CVE disclosure identifies an incomplete fix for CVE-2026-67278 in those initial v7 releases. Its complete v7 fixes are 7.24.3 and 7.23.6; that particular flaw is v7-only.
| Branch | Dated deployment reference | Why it matters |
|---|---|---|
| RouterOS 7 stable | 7.24.4 | Follows the completed security fix and corrects the affected-device LTE regression introduced in 7.24.3. |
| RouterOS 7 long-term | 7.23.7 | Follows the completed security fix and corrects the affected-device LTE regression introduced in 7.23.6. |
| RouterOS 6 long-term | 6.49.21 | The announced v6 fix for the affected issues; the v7-only CVE does not change this into a requirement to migrate immediately to v7. |
The vendor’s 7.24.4 announcement and 7.23.7 announcement identify the LTE hardware affected and the additional modem-firmware recovery steps for devices already damaged by the preceding builds. Review that procedure before changing a site whose only management path is LTE.
Do not write an inventory rule that approves everything “greater than 7.23.6.” That comparison would bless 7.24.2, which lacks the complete fix. Compare the relevant CVE against its affected range and branch. Then evaluate the supported deployment release for the actual hardware and features.
An urgent security update and a major-version migration are different projects. Combining them unnecessarily is an efficient way to turn one maintenance window into a weekend.
Check the subscriber side, too
Untrusted traffic is not limited to the public internet. A management address may be private and absent from external scans while remaining reachable from a customer VLAN. An old contractor VPN may reach far beyond its original purpose.
Review IPv4 and IPv6, translated and nonstandard ports, subscriber-facing interfaces, and the actual remote-management route. Test only paths you own or are authorized to assess. Moving SSH off port 22 changes the address on the door; it does not repair the lock.
Traffic to the router needs its own protection. Rules that filter only traffic forwarded through it do not establish a management boundary. Check effective rule order and interface membership, not just whether a rule named “protect management” exists.
The RouterOS service reference distinguishes allowed-address restrictions from firewall filtering. Use a deliberate firewall boundary and service restrictions as complementary controls. Check that the intended trusted path still works.
The short technical explanation
The actively exploited SSH chain combines faulty RSA-key authentication with improper privilege handling. Other disclosed issues involve bandwidth-test authentication and memory handling, SSH session state, RSA-signature validation, and WebFig authorization. Their prerequisites and affected versions differ; they are not six names for the same bug.
The operational lesson is broader than this vendor: an encrypted session and an authorized administrator are different states. Software can get the encryption right and the permission decision wrong. For exact findings, use the researcher’s affected-version table and disclosure, not a simplified fleet-wide assumption.
Contain exposure without cutting off your way back
Before disabling the service you are using, establish another trusted, tested management path and confirm recovery access. Check which backup, monitoring, and automation jobs depend on it. A remote tower is an inconvenient place to discover that the alternative login was theoretical.
These commands collect local information; they do not change the configuration:
/system resource print
/system package print
/ip service print detail
/tool bandwidth-server print
If they are genuinely unnecessary, individual services can be disabled:
/ip service disable ssh
/tool bandwidth-server set enabled=no
/ip service disable www
/ip service disable www-ssl
These are separate choices, not a paste-and-hope repair script. Disabling web services may interrupt REST integrations or other required web functions. If a service must remain, restrict it to necessary trusted sources and verify the result. Consult the bandwidth-test controls and service reference for the installed platform.
CERT Polska also recommends avoiding outbound TLS and built-in SSH-client operations on unpatched devices until updated. Review automated jobs as well as interactive use. Temporary restrictions reduce exposure; they do not replace the software fix.
Use a deployment sheet, not a memory test
For each device, record its role, running version, target version, exposure, owner, maintenance window, recovery path, and verification result. Include managed customer equipment and virtual RouterOS instances with real access, not just the impressive routers in the core.
- Prepare: Save recoverable configuration records privately. Verify architecture, required packages, storage, power, and a way back in.
- Obtain trusted software: Use official packages and a trusted acquisition path. Because outbound validation is implicated, a patched workstation and trusted transfer path may be appropriate. Do not turn off certificate checking to make a download succeed.
- Test a representative device: Cover meaningful hardware and feature differences. Shorten the cycle for active exploitation while containing exposure; do not substitute an arbitrary aging period for a risk decision.
- Deploy in controlled batches: Follow the vendor upgrade procedure. Confirm what actually booted.
- Accept the service: Check customer traffic, neighbors, tunnels, addressing, queues, management, logging, and configuration collection.
“It answers ping” is a useful first observation and an exceptionally small acceptance test. The deployment record should show a verified running version and service result, not merely that an update job was launched.
A patch does not erase an intrusion
Where supported, inspect the device-mode state:
/system device-mode print
A flagged device warrants investigation. An unflagged one is not thereby cleared. Read the flagged-state procedure before altering it; clearing the warning is not cleanup.
Compare against a trusted baseline: privileged users and keys, scripts, scheduler entries, DNS, firewall and NAT rules, tunnels, proxy settings, and unexpected files. MikroTik explicitly calls for configuration review after upgrading.
If there are credible intrusion signs, preserve useful logs and configuration, arrange service continuity where possible, and contain the affected equipment through a documented incident process. Avoid a hurried reset that destroys the evidence needed to understand scope.
A clean reinstall may be part of recovery; Netinstall is the vendor’s reinstall tool. Rebuild from a reviewed, trusted configuration rather than blindly restoring the suspect backup. Replace affected credentials and keys from a trusted system, including reused secrets elsewhere. A freshly installed router with a stolen password has received only half a repair.
Keep the resulting version inventory, configuration history, and verification record. They turn the next advisory from an archaeological expedition into a manageable engineering job.
Oaktown Tech can help assess management exposure, plan a staged remediation, and verify network operation afterward. Bring the device roles, version inventory, and symptoms to Start a Project; keep credentials and sensitive exports out of the inquiry form.

Join the discussion