The Gaia OS DynamicCLI daemon runs as root and listens on a local socket at /tmp/dynamic_socket. On the affected builds that socket exists at 0777 with no authentication of the connecting process, and the daemon executes whatever command string it receives as root 1.

That sentence covers the whole vulnerability. Check Point rates CVE-2026-123039 at 7.8, and the vector begins AV:L, so the attacker needs code execution on the appliance before the socket matters 1. NVD’s record still opens in the Received state, with no base score of its own, which leaves 7.8 as the vendor’s number 1.

The permission is the bug

I went looking for a memory bug and found a file mode. A root command interpreter sat behind a filesystem permission, and the permission was 0777, so the boundary between a portal worker and a root shell came down to a mode bit 2.

The fix logs the connections it refuses, and the example Check Point publishes uses uid 99, the nobody account 2. The old mode let a process with no rights of its own write one string to a socket and get root.

On a firewall this turns one compromised service into one compromised appliance. On a management server it turns a compromised service into the source of the policy every gateway underneath enforces. I rank the second case higher, because more hosts sit behind that root shell.

How the October intrusions used it

Check Point Research published the chain on 12 November. Scanning and exploitation from one address began on 30 October against Mobile Access portals, with PHP2Shell, CVE-2026-109618 and CVE-2026-109619, as the way in. The portal service account was the foothold, and CVE-2026-123039 is how that account reached root 3.

From root on the gateway the attackers wrote PHP webshells into the portal’s template directory, ran a backdoor disguised as CPconfigdaemon, dropped a shared library at /lib/libcpfwcommon.so that intercepts networking calls inside appliance processes, and scanned the internal network with gogo. They then moved to a VMware ESXi host over SSH as the vpxuser service account, ran the destroy_v6 wiper, and left a note named DREAM_ANNEXATION.txt in the directories it processed 3.

Check Point says the Gateway and Management fixes were available on 23 October, a week before the first intrusion it documents, and that the compromised population came to a couple of dozen unpatched systems in one region 4. The backdoor and the library it drops are modified variants of PhiliKit, a framework ESET tied to China-nexus activity under the name UNC5221, and the research post says the evidence does not allow a confident assessment of who ran this operation 3. I will leave attribution there, because the part I can act on is which builds carry the fix and which ones sit on a mitigation.

The fix line stops at R81.20

Check Point lists the fixed builds as Jumbo Hotfix takes: R82.20 from Take 14, R82.10 from Take 52, R82 from Take 132, and R81.20 from Take 174 2. NVD’s affected table puts the last vulnerable builds at Take 10, 50, 130 and 173 on those four versions 1. The two lists leave a one-take gap on R82.10 and on R82, at Takes 51 and 131, and R82.20 opens to a three-take gap across Takes 11 through 13; R81.20 runs straight from 173 to 174. On the other three, an admin who assumes the lists meet lands a take short of the fix, so I read the take number off the appliance instead of inferring it from the version string.

The version list on the advisory carries R81 and R81.10 as well, both marked EOS, and neither appears in the fixed-build list 2. Check Point recommends installing the November Jumbo Hotfix on all management servers and security gateways 4, and no take of that hotfix covers either EOS release, so an admin on R81 or R81.10 either runs the mitigation or moves the box to a release that carries the fix. The mitigation is dbset process:dynamic_server, then dbset :save, and Check Point notes that stopping the process breaks the Dynamic CLI commands in sk144112, which is a real trade if anyone uses Dynamic CLI there 2.

A fixed build leaves /tmp/dynamic_socket at srw-rw---- root config, mode 0660. ls -l /tmp/dynamic_socket settles whether the build on the box carries the fix, and I would run that rather than read a version label.

Reading the log after the patch

The daemon writes every command it receives to /var/log/dynamicCli.log, and legitimate traffic arrives inside a fixed Clish wrapper that sources CPprofile.sh and sets SP_DISTRIBUTE=0 and noglob before the Gaia binary runs 2. A line with Received = and no wrapper is a root command that came in through the socket, which is the thing to hunt on a host that sat reachable before the fix.

Check Point’s greps cover three cases: commands outside the wrapper, the #return_value# marker left by its system() sink, and a string list from exploitation attempts that includes whoami, /etc/shadow, ncat, /dev/tcp and authorized_keys 2. A patched box gains Rejected connection from unauthorized uid lines instead, and a burst of those wants correlating against the portal and authentication logs for the same window 2.

The log tells you someone used the socket. It does not tell you what the root shell read, and that gap stays open after the patch. The KEV entry lists forensic triage alongside the patch date, and the two are different jobs 5.

The trust boundary question

A management plane carries an odd shape: a handful of hosts with a large amount of authority, and a local account population nobody counts as privileged. This bug needs one local account. On a gateway that account is a portal service. On a management server it could be a backup agent, or a session someone left open on a jump host.

When I look at a box that pushes policy downward, the question I ask is what a low-privilege process on it can reach, and whether the answer changes between the appliance and the console that configures it. I get the same 7.8 and the same AV:L vector on either one, so I decide the difference from what sits behind the root shell.

CISA added this to the KEV catalog on 13 November with a due date of 16 November, and the row covers gateways and management servers together 5. A gateway and a management server carry different weight, so I decide which one my environment moves first rather than working from the catalog order.


  1. National Vulnerability Database. (2026, November 12). CVE-2026-123039 detail. National Institute of Standards and Technology. https://nvd.nist.gov/vuln/detail/CVE-2026-123039 ↩︎ ↩︎ ↩︎ ↩︎

  2. Check Point Software Technologies. (2026, November 12). CVE-2026-123039: Local privilege escalation to root in the DynamicCLI process via local world-writable control socket (Solution ID sk1000362). https://support.checkpoint.com/results/sk/sk1000362 ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎

  3. Check Point Research. (2026, November 12). PHP2Shell exploitation on Check Point gateway. https://research.checkpoint.com/2026/php2shell-exploitation-on-check-point-gateway/ ↩︎ ↩︎ ↩︎

  4. Check Point Software Technologies. (2026, November). Security advisory: Active exploitation of PHP2Shell and local privilege escalation. Check Point Blog. https://blog.checkpoint.com/security/security-advisory-active-exploitation-of-php2shell-and-local-privilege-escalation/ ↩︎ ↩︎

  5. Cybersecurity and Infrastructure Security Agency. (2026, November 13). Known exploited vulnerabilities catalog (Catalog version 2026.11.13). https://www.cisa.gov/known-exploited-vulnerabilities-catalog ↩︎ ↩︎