Skip to content
COR Intelligence
Back to The Patch Wave
The Patch WaveAct todayFound by: not stated

Web hosting servers: poisoned cPGuard security update stole server passwords and keys

COR Solutions4 November 2026
6 min read1,333 words

An attacker broke into OPSSHIELD and pushed a fake cPGuard update to about 2,950 hosting servers on 02/11/2026. It copied root password hashes, SSH keys and database passwords. Ask your host if it uses cPGuard.

Keys hang from the lock of an open server rack door, with green and amber server lights and a closed laptop behind them in a bright hosting room.

What happened

cPGuard is a security add-on that many smaller web hosts run on their Linux servers. OPSSHIELD, the company behind it, says an attacker signed up as a customer under a false identity, broke into its billing and licensing server, took the keys it uses to publish updates and pushed a fake cPGuard update. Between 02:08 and 03:20 UTC (UK time) on 02/11/2026, about 2,950 customer servers installed it. The fake update copied the server's password hashes, root SSH keys, the MySQL root password and shell history to the attacker. OPSSHIELD says it took about ten hours from the attacker's first probe of its billing portal to the fake update going out.

OPSSHIELD also says the attacker took a copy of about 7,900 billing portal accounts: name, company, postal address, email address and password. The passwords were stored as bcrypt hashes, not as plain text, and OPSSHIELD has since reset every one.

Does this affect you

Businesses whose website or email sits on a server that runs cPGuard, and anyone who manages their own hosting server with it. OPSSHIELD says cPGuard is for hosting companies that run VPS or shared hosting servers, and for small businesses that own a server. Your host or web developer will know.

Go by the version check below, not the clock. OPSSHIELD's disclosure says a server that didn't have cPGuard online in that window did not get the fake update, but its earlier advisory shows example installs of the bad versions as late as 12:46 the same day.

OPSSHIELD lists only system files as taken and doesn't say website files or mailboxes were copied. It says it hasn't found other malicious activity so far, and calls that finding preliminary.

What to do

  1. Ask your host or developer today: "Do any of our servers run cPGuard, and did any install versions 5.88.12 to 5.88.19?" They can check the server's cPGuard install log with the command in OPSSHIELD's advisory.
  2. If yes, ask them to update cPGuard to the latest version (5.88.26 when we checked on 05/11/2026), then check by hand that the three files and folders OPSSHIELD lists are gone, and delete them if not. OPSSHIELD says a follow-up update removes the last of them, but its changelog doesn't list that update yet.
  3. Then change the server's root and user passwords, replace its root SSH keys (and remove the old ones from every server that trusted them), change the MySQL passwords, and change any password or key that appears in the server's shell history. OPSSHIELD says this takes most admins under an hour per server.
  4. Treat any other server that this one can log in to with SSH keys as exposed until its keys are replaced.
  5. If you or your developer have an OPSSHIELD login: its login page says every account password has been reset. Set a new one with the reset option, then remove and re-add two-factor login.
  6. If OPSSHIELD told you the attacker logged in to your server by hand, rebuild it from a clean backup taken before 01/11/2026. OPSSHIELD says it will walk customers through the steps at no charge.
  7. If your customers' personal data was on an affected server, ask your host in writing what was on it. The ICO says a personal data breach that meets its threshold must be reported to it within 72 hours, and the clock starts when you discover the breach.

Is it being attacked?

Yes. This was a real attack, confirmed by OPSSHIELD itself on 03/11/2026. In its first advisory OPSSHIELD estimated the attacker had retrieved collected files from 32 customer servers, and said that was a preliminary figure for confirmed retrieval, not the total number of servers that may be affected. Its later disclosure tells every customer whose server ran the fake update to assume its root-level secrets are known to the attacker.

Found by

OPSSHIELD says it found and stopped the attack itself. It names no people and doesn't say how it spotted the fake update. Its timeline shows cPGuard flagging and banning the attacker twice on the evening of 01/11/2026, before the break-in. Tag: not stated.

For developers

  • No CVE. Fake update scripts cpguard-v58812-i91511*.sh (four variants, SHA-256 hashes in the disclosure), shipped as cPGuard 5.88.12 to 5.88.19, ran as root. Timeline (UTC, disclosure): fake update uploaded 02/11 02:04 to 02:09 with a stolen update-storage access key; installed by agents 02:08 to 03:20. OPSSHIELD says its logs don't record the exact request that gave the attacker access.
  • Collected: /etc/shadow, /etc/passwd, /root/.ssh, /root/.my.cnf, root shell history, and /root/.s3cfg on a small number of servers. Left a setuid root shell at /var/tmp/.cpg91511_sh, files in /var/tmp/.i93677/ and a scheduled task at /etc/cron.d/i93677. The first advisory also lists the staging folder /var/tmp/.cpg91511/ and the archive /var/tmp/.cpg91511/loot.tgz, and says their absence after clean-up doesn't show the script never ran.
  • Check install history: grep 'Successfully updated' /opt/cpguard/logs/install_log. OPSSHIELD says an entry for any version from 5.88.12 to 5.88.19 means treat the server as potentially compromised even if it now runs a clean version, and that no matching entry, or a missing or incomplete log, doesn't prove it was safe.
  • Update: cpgcli update --start, then cpgcli --version (advisory). The disclosure says 5.88.26 removes the root shell and a follow-up removes the scheduled task. The earlier advisory says 5.88.25 or later removes the backdoor and staged files. The changelog on 05/11/2026 lists 5.88.25 ("Patched compromised update") and 5.88.26 ("Revoked support access to servers if it is enabled"), both dated 02/11/2026, and nothing newer, so check /etc/cron.d/i93677, /var/tmp/.i93677/ and /var/tmp/.cpg91511_sh by hand.
  • Rotate: root and every local user password (the advisory says from a trusted recovery environment, and to change those passwords anywhere else they were reused), root SSH keys (remove the old public keys from authorized_keys on every server that trusted them, and review authorized_keys for anything you don't recognise), MySQL and other database passwords, anything in root's shell history, S3 keys and DigitalOcean tokens.
  • OPSSHIELD's own clean-up: removing its support-access keys from cPGuard servers where they had been added (advisory, 03/11/2026), re-issuing all cPGuard licence keys so the copies the attacker holds stop working, and resetting every billing portal password (disclosure and portal login page). It isn't recommending a routine operating-system rebuild, and calls that view provisional (advisory).
  • IoC: operator IP 94.183.188.91, which sometimes sent a fake cPGuard/5.88.11 user agent; web root probe files and web shells following an i9xxxx pattern; *_rx.php endpoints used to receive stolen data. The disclosure also lists a gsocket backdoor and a Go implant found on OPSSHIELD's own internal server, not on customer servers, and says to check from a trusted boot environment because the implant hides itself from standard tools.
  • OPSSHIELD's view, not evidence: "Attackers now use AI tools to scan for weak points, read unfamiliar code and write custom tools in hours instead of weeks; the speed of this attack, from first probe to fake update in about ten hours, reflects that." (disclosure, 03/11/2026)
  • Not a CISA KEV item (no CVE; catalogue 2026.11.04, released 04/11/2026, checked 05/11/2026).

Sources

Read next

More from The Patch Wave