MikroTik RouterOS takeover chain hits devices with exposed SSH

CERT Polska published a coordinated disclosure on 5 September 2026 covering six RouterOS vulnerabilities. Two of them form a chain the team named MikroTrick: unauthenticated full admin control when SSH is reachable from the internet. MikroTik posted a 4 September forum security update (normis).

Fixes are already in 7.25beta3, 7.24.2, 7.23.4, and 6.49.21. The chain is not in the CISA Known Exploited Vulnerabilities catalog.

CVE-2026-67276 (CVSS 9.2) is an SSH public-key check that compared RSA type and modulus but not the full key and exponent. An attacker who knows a username and that modulus can craft a key and log in without the private key. CVE-2026-86060 (CVSS 9.2) lets a username that begins with a disallowed character, logged as -2, reach a legacy helper and escalate to full admin. A third bug, CVE-2026-67277 (CVSS 8.8), is bandwidth-test memory disclosure and denial of service.

CERT says exploitation is confirmed against devices whose SSH service is on a public network, not against every home RouterOS box sitting behind the default firewall. MikroTik told home users that default configs are not at immediate risk and still told everyone to upgrade. In the same thread, normis said many years of versions are affected and that the fix exists only from the named releases on.

Successful attacks, including creation of a privileged user named ops, have come from 82.192.72.4 since at least 2 September. CERT also logged exploit attempts from 103.102.31.18. The logged lines are login failure for user -2 from <ip> via ssh and user <name> added by ssh:-2@<ip>.

Patched builds stop the observed attacks. After the upgrade, RouterOS can set a Flagged device-mode marker when it sees known compromise fingerprints. Absence of Flagged is not proof the box is clean.

CERT used GPT-5.5-cyber and GPT-5.6-sol inside a supervised GTAC lab to speed hypothesis search. Every finding was then verified on real RouterOS by researchers, with negative controls and clean-state repeats. The work was human-supervised lab research, not a model that found the bugs on its own. That lab access sat under OpenAI's government program, adjacent to the company's Daybreak frontline-defender pledge.

Some offers on this page may be paid placements or contain affiliate links.