Detection engineering RouterOS / SSH

RouterOS rekey bypass: what your SSH monitoring can tell you

An encrypted connection does not establish who is using it. CVE-2026-67279 makes that distinction an operational problem: the SSH server can accept session activity before a user has authenticated.

CISA added this flaw to KEV on September 25. Its entry identifies a possible chain to CVE-2026-86060 and lists September 28 as the due date for covered federal systems under BOD 26-04. Our earlier MikroTrick article covers the wider disclosure. This follow-up looks at the monitoring problem the rekey path exposes. CISA KEV feed

01 / Transport is not identity

Refreshing encryption must not advance a user's privileges

SSH establishes encrypted transport before user authentication. Rekeying refreshes that transport's keys. It does not authenticate a user or authorize a session. In affected RouterOS builds, a client-requested rekey crosses that boundary anyway: the server accepts session-channel and execution requests without a completed login. SSH transport specification · CERT Polska advisory

Same starting condition / No user authenticatedRekey changes the encryption. It must not change the permission.

Expected handling

  1. TransportKeys refreshed

    Encrypted connection

  2. AuthenticationNo user authenticated

    Gate stays closed

  3. Session requestChannel rejected

    Login is still required

Affected RouterOS

  1. TransportKeys refreshed

    Encrypted connection

  2. AuthenticationNo user authenticated

    State check bypassed

  3. Session requestChannel accepted

    Exec requests reachable

Separate escalation step

CVE-2026-67279 opens unauthenticated session access. CVE-2026-86060 can then turn attacker-supplied login data into administrative privileges.

Conceptual state comparison, not a captured session.

The two flaws should not be collapsed into one. Bishop Fox reproduced the combined administrative takeover on vulnerable 7.x builds. Its tested 6.x construction reproduced the rekey bypass without obtaining the same administrative session. That is a limit of that test, not a reason to leave affected 6.x devices unpatched. Bishop Fox research

02 / Match the record to the question

Three views of the same incident

I would ask the network team and the router owner to compare what they actually retain. A connection record, an authentication decision, and a configuration change answer different questions. None should stand in for the others.

01 / Network

Who reached SSH?

Flow records and passive capture

Useful evidence
Endpoints, connection time, duration, and traffic volume.
Limit
Cannot directly read encrypted authentication, rekey, or session messages.

Establish contact and timing.

02 / SSH endpoint

What did the server allow?

Protocol-aware instrumentation

Useful evidence
Authentication state and accepted or rejected channel and exec requests.
Limit
Requires explicit instrumentation. Ordinary login logs may not expose these decisions.

Establish the state transition.

03 / Device

What changed or remained?

Configuration and audit history

Useful evidence
Accounts, permissions, scripts, scheduled tasks, and tunnels.
Limit
A change needs context. It does not identify the entry vulnerability by itself.

Establish impact and persistence.

Rekey messages themselves travel under the existing encryption. A passive sensor may identify SSH and describe the connection, but it cannot simply read the rekey-to-channel sequence from that encrypted traffic. That is why a rule matching an early rekey is neither a general network signature nor proof of this bypass. RFC 4253, section 9

03 / Look beyond login history

The router's configuration may outlast its logs

Bishop Fox found persistence in a script and scheduled task that recreated a privileged account. The relevant login history was no longer available. Its report also highlights unusual numeric object ownership as a hunting lead, while noting that legitimate objects can have unfamiliar owners. The useful test is whether you can explain the object, its permissions, and its behavior. Field investigation

  1. Establish exposure and preserve what remains

    Record the installed build and management-access paths. Preserve available remote logs and a protected configuration export before resetting the router. Use the actual exposure period, not the KEV listing date, to set the review window.

    Keep: build, exposure history, original exports
  2. Account for privileged changes

    Compare users, scripts, schedulers, proxies, and tunnels with an approved baseline and change records. Review what a scheduled task does, not just whether its name looks familiar. Check RouterOS's Flagged state alongside that review.

    Keep: object, owner, behavior, explanation
  3. Investigate indicators in context

    CERT Polska reports the ops account and SSH log entries involving -2. A failed login or source-address match is a lead. An unexplained privileged account or account-recreation task warrants a different level of response.

    Keep: correlated events and findings

If compromise is suspected, coordinate isolation and recovery with the network or operations owner. Preserve evidence before resetting the device, rebuild from a verified configuration, and rotate the secrets at risk. CERT Polska warns that the absence of a Flagged marker does not establish that the router is clean. Investigation and recovery guidance

04 / Test the coverage you intend to claim

A believable SSH prompt is not protocol-level exploit coverage

For a deception deployment, the acceptance test should follow the intended use. Recording password guesses and submitted commands is valuable. Observing a pre-authentication state error requires visibility below the command prompt.

A RouterOS-looking banner does not reproduce RouterOS's SSH state machine. A decoy intended to study this mechanism needs controlled validation of what it observes, rejects, or deliberately emulates. Keep that work isolated from production and verify the event-to-alert path with the team that will investigate it.

Ask the monitoring team to show the evidence behind the alert. A connection, an accepted session, and an unauthorized configuration change are different findings.

05 / Close the exposure

Use the fix boundary to assess exposure, not to pick an old build

CERT Polska identifies 6.49.21, 7.23.4, and 7.24.2 as fixed releases for CVE-2026-67279. Related CVE-2026-67278 needed later fixes in 7.23.6 and 7.24.3. The first set therefore does not close every issue in the wider disclosure. CVE-specific fix details

Install a current supported release for the appropriate channel after checking device-specific release notes. Restrict SSH to approved management paths while doing so. MikroTik's advisory recommends reviewing configuration even after the update. Current release notes · Vendor security guidance

Release-channel check, September 25

The vendor currently lists 7.24.4 stable, 7.23.7 long-term, and 6.49.22 long-term. The later 7.x builds address an LTE regression introduced in 7.24.3 and 7.23.6. Confirm the channel and hardware requirements before scheduling the change. These are dated release observations, not a substitute for checking the current changelog.

Sources and review

Reviewed September 25, 2026. The vulnerability and reproduction findings are attributed above. The monitoring checklist is our analysis of the evidence those findings call for.