Skip to content

Build1 publisher2 min readPublished

Firewalling RouterOS SSH to a management network removes MikroTrick's precondition

CERT Polska confirmed exploitation of the MikroTrick chain on September 5, and the two bugs need no password and no private key, only a reachable SSH service, so the firewall rule scopes the risk before any build audit can.

The Engineer · Build desk

Illustration accompanying Firewalling RouterOS SSH to a management network removes MikroTrick's precondition

What happened

  • CERT Polska disclosed the MikroTrick chain in MikroTik RouterOS on September 5, 2026, and confirmed that it is being exploited.
  • The chain pairs CVE-2026-67276, missing authentication in the SSH public-key verification path, with CVE-2026-86060, an argument delimiter injection that escalates the session to administrator.
  • Neither bug needs a password or a private key, and the only stated precondition is network reachability to the SSH service.
  • ZoomEye queries run on September 19, 2026 return totals of matching index records. A match does not confirm a reachable SSH service or a vulnerable version.

Compiled by The EngineerSomething wrong?How this is made

Why it matters

  • constraint An estate with SSH bound to a management network is outside this chain's reach whatever build it runs. That firewall change is verifiable in a way a version census across a field fleet is not.
  • decision Exposure counts decide where to look. Every candidate device still needs asset-level confirmation before anyone marks it clean.
  • exposure Shrinking the reachable surface does not undo an intrusion that already happened, so anything that was answering SSH needs the log check for a -2 login and an audit of local accounts.
  • cost A device that reports itself as Flagged costs a rebuild with evidence preserved, and that labour lands on whoever touches the hardware.

Reachability is the gate. The chain starts in the SSH public-key verification path, and the dev.to writeup puts the consequence plainly: a RouterOS device that is not reachable over SSH from the internet cannot be attacked through this chain, regardless of its software version [5]. The statement is about the transport, so it holds for every build. Restrict SSH, WebFig and the bandwidth-test service to a trusted management network and the precondition is gone [6].

The writeup advises checking logs for SSH logins using the username `-2`, and auditing local users for unauthorized accounts such as `ops` [16]. CVE-2026-86060 is an argument delimiter injection [3]. A login name that starts with a dash parses as an option once it reaches a command line, which fits that bug class [21]. The log audit is a string search for a username that is really a flag.

On the measurement side, the ZoomEye data was collected on September 19, 2026, fourteen days after the advisory [19]. The writeup is explicit about the limits. Fingerprint matching is not version detection, so a record matching `app="RouterOS"` shows that the service identifies as RouterOS and not whether it runs the patched 7.24.2 or an older release [11]. The totals describe the ZoomEye index, which samples and retains what it has observed, not a complete census [12]. And the port 8291 count is Winbox, a different management surface from SSH [8]. The article lists its query strings but does not include the numbers they returned [10].

The fixed versions are 6.49.21, 7.23.4, 7.24.2 and 7.25beta3, and 7.23.5 was released on September 4, a day before the disclosure [14][20]. An inventory check therefore has to match five build numbers across two major trains, 6.x and 7.x [22]. Upgrading also installs a detector: patched builds add a `Flagged` self-check that marks a device if it finds known tampering artifacts on boot [15].

That leaves the population number. For the widely cited 122,000 to mean anything about a specific estate, the counted devices would have to be ones you own, and SSH would have to answer from outside. No fingerprint query establishes the second condition [9][11]. The writeup attributes the figure to public reporting and describes it as exposed and potentially vulnerable instances. It does not present them as confirmed intrusions [13].

What to watch

  • A version-level breakdown of the exposed RouterOS population would show how much the firewall change buys. This data set did not attempt one.
  • Counts of devices reporting themselves as Flagged, if CERT Polska or MikroTik publish them, would separate exposure from intrusion.
  • Any report of the chain reaching devices through Winbox on port 8291 would widen the precondition beyond SSH.
Loading claim ledger
Loading source directory links
Loading share composer
Loading topic controls
Loading related stories