Vulnerabilities
A world-writable socket gave root on Check Point firewalls
CVE-2026-123039 lets a local user run commands as root on Check Point gateways. Attackers chained it from a web shell to a wiper. What to patch and check.
Check Point has confirmed attackers are chaining two of its flaws to get from the public internet to root on its firewalls, and the United States Cybersecurity and Infrastructure Security Agency (CISA) added the second link in that chain to its Known Exploited Vulnerabilities (KEV) catalog on November 13, 2026. CVE-2026-123039 carries a CVSS score of 7.8 and lets any authenticated local user on a Check Point appliance run commands as root. On its own it is a second-stage flaw, because it needs code already running on the box. Attackers had that from a web shell on the Mobile Access portal, so the two flaws together hand over the whole gateway.
This affects Gaia-based Check Point gateways, Security Management and Log servers, Multi-Domain Management and Log servers, and Maestro orchestrators running R82.20, R82.10, R82 or R81.20, plus the end-of-support branches R81.10 and R81. Spark Firewalls and Smart-1 Cloud are not affected. Federal agencies have until November 16 to fix it. Check Point says it has seen exploitation against a couple of dozen customers, and in some intrusions the attackers went on to wipe files on a connected VMware ESXi host. Below is how the flaw works, what the intrusions looked like, and the exact builds and checks you need.
How it works
Gaia, the operating system on Check Point appliances, has an extended command line interface (CLI) served by a background process called dynamic_server. That process runs as root and listens on a local socket at /tmp/dynamic_socket. A socket is an interprocess channel, in this case a file other programs write commands to.
Two things go wrong at once. On the affected builds the socket is created world-writable, mode 0777, which shows as srwxrwxrwx in a directory listing, so any local user can write to it. And the daemon never checks which user is on the other end before it acts, so it runs whatever command string arrives as root. That is why CISA tags the entry with both CWE-732 (incorrect permission assignment for a critical resource) and CWE-862 (missing authorization). The CVSS vector is AV:L/AC:L/PR:L/UI:N, meaning local access and low privileges are enough and no user interaction is needed.
The entry point was a separate pair of flaws Check Point calls PHP2Shell, tracked as CVE-2026-109618 and CVE-2026-109619, in the Mobile Access portal that remote workers log in through. PHP2Shell gave the attacker code execution as the low-privileged portal service account. That account could reach the socket, so a single write to /tmp/dynamic_socket turned a constrained web shell into full root on the firewall. A pre-authentication web flaw and a local escalation are each serious; on the same appliance they compose into remote code execution as root.
What attackers are doing
Check Point Research tracked two waves. The first began on October 30, 2026 against Israeli organizations and included destructive activity. A second, broader wave on November 2 hit organizations in several countries and looked opportunistic, with no destruction observed. Both came about a week after a fix for the PHP2Shell flaws was already available, and much of the traffic came from 130.94.19.204.
After reaching root, the attackers settled in. They dropped PHP web shells named zz*.tpl into the portal's templates directory, then installed a passive Linux backdoor Check Point calls TSGA. It launches from /tmp/CPconfigdaemon, relocates itself to a boot-persistence directory as slot_CPconfigdaemon, listens on TCP port 11072, and can tamper with the binary firewall logs on the device. A companion library, TSProxy, is written to /lib/libcpfwcommon.so and injected into appliance processes to hide the attacker's connections. A reconnaissance tool called gogo was run on the gateway to map the internal network.
In the destructive cases the firewall was only a foothold. The attackers used the stolen position to reach a VMware ESXi host over SSH with the vpxuser service account, then ran a Linux wiper named destroy_v6 that overwrites files with zeros before deleting them and leaves a note called DREAM_ANNEXATION.txt in each directory it touches. Check Point notes that TSGA and TSProxy belong to the PhiliKit toolset that ESET has linked to China-nexus activity, but says the wiper and the Israel focus do not fit that group cleanly, so it holds attribution at low confidence.
What to do
Install the November Jumbo Hotfix for your branch, which Check Point ships in advisory sk1000362 and which rejects and logs unauthenticated calls to the socket. The fixed takes reported for this flaw are R82.20 Take 14, R82.10 Take 52, R82 Take 132 and R81.20 Take 174 or later. R81.10 and R81 are past end of support and have no fix, so plan an upgrade rather than a patch. Take numbers for a given branch move, so confirm the exact take for your version in sk1000362 before you deploy; at least one tracker still showed no published fix, so do not assume you are covered until you have checked.
If you cannot patch today, stop the daemon. In Expert mode run dbset process:dynamic_server f then dbset :save, which stops dynamic_server and keeps it from restarting. This disables the dynamic CLI commands listed in sk144112, so reverse it with dbset process:dynamic_server t after you patch.
To check whether you were already hit, start with the ones facing the internet. Confirm the installed take with show version all in clish or cpinfo -y all in Expert mode. Look at the socket itself with ls -l /tmp/dynamic_socket: a world-writable srwxrwxrwx means the box is still unpatched, while a fixed appliance should show srw-rw---- owned by root. On a patched box, grep "Rejected connection from unauthorized uid" /var/log/dynamicCli.log lists blocked escalation attempts, and any hit is worth investigating. Also review that log for commands that did not come through the normal Clish wrapper. Then hunt the post-exploitation artifacts: zz*.tpl files in the portal templates directory, the files /tmp/CPconfigdaemon and /lib/libcpfwcommon.so, a process listening on TCP 11072, and SSH logins to any ESXi host as vpxuser from a gateway address.
Treat a gateway that was exposed and unpatched as compromised until the checks come back clean. Patching does not evict an attacker who is already inside, so rotate every credential the appliance stored or passed through, including VPN and management secrets, review local accounts and authorized_keys files for additions, and rebuild from a known-good image where the implants above appear. The ESXi side needs its own review: rotate the vpxuser and host credentials and restore wiped datastores from backups.
The wider lesson
A firewall is a Linux server, and a management socket left world-writable is the kind of local flaw that sits unnoticed until a pre-authentication bug on the same box makes it reachable. On an appliance that terminates remote-access VPN, the gap between a service-account web shell and root should be a hard wall, not an open file in /tmp. Treat appliance operating systems like any other host you harden: least privilege on local sockets and files, and a patch path that closes second-stage flaws as fast as the headline remote ones.
If you want an outside check on which of your gateways and portals are actually reachable and what an attacker could chain once inside, our network penetration testing and safeguarding and hardening work covers exactly this. Open the chat and Yaali, our AI agent, will pass your question to the engineer who would do the work.
Sources: Check Point Research: PHP2Shell exploitation on Check Point gateway, The Hacker News, CISA KEV catalog, safeguard.sh, NFLO knowledge base, Canadian Centre for Cyber Security AV26-1154.