The Firewall Was the Way In
NOVEMBER 17, 2026 · UMERCS

A Check Point security gateway is the box that decides what gets in and out of a company. Firewall policy, VPN, remote access, logging, the whole perimeter. Between 30 October and 4 November 2026, attackers did not go around that box. They went through it, and then used it against the network standing behind it.
Check Point confirmed active exploitation on 12 November against a chain of two flaws: PHP2Shell, tracked as CVE-2026-109618 and CVE-2026-109619, in the Mobile Access portal, and CVE-2026-123039, a local privilege escalation in Gaia OS that turns a web shell into root. CISA added CVE-2026-123039 to the Known Exploited Vulnerabilities catalog on 13 November, with a federal remediation deadline of 16 November. That is seven days from confirmation to deadline, which is a pace most organisations do not have.
The chain, in the order the attacker walked it
The chain is short and it reads like a plan. Break into the internet-facing portal with PHP2Shell. Land as the portal service account. Escalate to root through the Gaia DynamicCLI socket. Install a backdoor, keep a tunnel, and from the appliance reach whatever the appliance can reach.
Check Point Research describes two waves. The first, on 30 October, targeted Israeli organisations and included destructive activity against some of them. A broader second wave on 2 November hit organisations in multiple countries and sectors with no evidence of destructive activity in the investigated cases. Exploitation was observed until 4 November.
The appliance at the edge is a specific kind of target. It holds the VPN configuration, the firewall rules and the trust relationships that let a management server push policy to every gateway. Compromise it and you inherit all of that, in place, using credentials and routes that were already approved.
Step one: the door everyone already knew about
PHP2Shell landed on 21 October, when the PHP project shipped an out-of-cycle release for two unauthenticated remote code execution flaws in core. Anything serving PHP on an exposed interface needed the update within the week. A Check Point gateway running the Mobile Access portal is exactly that.
Check Point scored the portal flaw 9.2 and published guidance in sk1000302. A hotfix for it was available on 23 October. The first observed exploitation was on 30 October.
Seven days. That is the entire gap between a fix sitting on a support site and an attacker on the appliance. The organisations that were hit were not defeated by clever research. They were beaten by a maintenance window that had not arrived yet.
The part scanners miss
A severity list sorts by score, and the score is a property of the flaw in isolation. Two flaws at 7.8 are worse than one at 9.2 when the first one is how the attacker gets into the position the second one needs. Any patch process keyed to a threshold, patch everything above 8.0, will let a chain like this one sit for as long as the calendar allows.
The practical version of that argument is a short list of questions, not a rank. Which of my exposed systems can be reached by someone who is not supposed to reach them. What does a foothold on that system unlock. And who is accountable for the answer. A 7.8 that turns a web shell into root on the box controlling your VPN belongs at the top of the queue, ahead of a 9.8 in an application nobody exposes.
Check Point's own timeline shows how quickly that gap closes. Disclosure on 21 October. Hotfix on 23 October. First exploitation on 30 October. Second wave on 2 November. The window for a vendor-driven, needs-a-project-plan fix was gone before most organisations had approved the change.
What the attacker inherits
There is a reason edge appliances keep showing up in these chains, and it is not laziness. The appliance is a trust anchor. Its configuration is the map of what your network permits. It holds certificates, routes, VPN profiles and often an LDAP or RADIUS binding that lets it decide who is allowed in. A shell on that box is not one compromised host. It is a position inside the mechanism your business uses to keep people out.
That also changes what remediation costs. Rebuilding a compromised laptop is an afternoon. Rebuilding a gateway means treating the VPN, the firewall policy and everything that trusted the appliance as potentially touched. Check Point's guidance reflects that: it points operators to log review, published indicators and, where the appliance was exposed and unpatched in the exploitation window, an investigation rather than a reboot.
Attribution is the open question, and Check Point's language is careful. The toolset overlaps with activity previously seen against Ivanti appliances, and the company states the evidence does not support a high-confidence conclusion about the operation. That uncertainty is a reason to look at your own logs sooner, not a reason to wait for the picture to clarify.
Step two: one socket, no questions
The second flaw is the interesting one, because it is not exotic. On affected Gaia versions, the root-owned dynamic_server process listens on a local socket at /tmp/dynamic_socket. The socket is created world-writable, mode 0777, and the process does not authenticate whoever connects to it. Any authenticated local user can write a single request to that socket and have it executed as root.
Check Point classes it as CWE-862, missing authorization, and CWE-732, incorrect permission assignment for a critical resource. CVSS 7.8, local only, which is why it looks unremarkable in a scanner report. You need code execution on the appliance first. PHP2Shell supplied exactly that.
Read those two flaws together and the shape of the problem is clear. A 9.2 remote code execution flaw gets a project plan, a change request and a maintenance window. A 7.8 local escalation gets a footnote. Chained, the footnote is the one that hands over root.
The fix is the November Jumbo Hotfix Accumulator, and the Takes matter: R82.20 Take 14 or later, R82.10 Take 52 or later, R82 Take 132 or later, R81.20 Take 174 or later. R81.10 and R81 are end of support and carry no fix. The same Takes also close CVE-2026-107484, an authorization flaw in Gaia Portal backend handlers scored 7.3 that Check Point lists as not exploited.
Step three: staying, and one case of destruction
Check Point Research reports PHP webshells on the compromised appliances, plus a backdoor it calls TSGA carrying an embedded networking component named TSProxy. In the investigated intrusion, the attackers moved from the gateway to a VMware ESXi host over SSH and ran a wiper referred to as destroy_v6. Check Point says destructive activity happened in a limited number of cases.
Attribution is not settled, and Check Point says so plainly. The company links TSGA and TSProxy to PhiliKit, a toolset previously reported in attacks on Ivanti appliances, while stating the evidence does not allow a high-confidence assessment of the operation's true nature.
One detail is worth sitting with. The compromise left a ransom-style note referencing Dream Annexation with no payment demand attached. Destruction without a business model is a different kind of problem than extortion. You cannot negotiate your way out of it, and you cannot assume the attacker's incentive is to keep your systems running.
Why the week matters more than the bug count
Check Point's own figures are small and that is the point. A couple of dozen unpatched systems, in a single region, were compromised in the observed incidents. The company also notes that the PHP2Shell hotfix had been available a week ahead of the attacks.
A couple of dozen is not a mass exploitation event. It is a demonstration of what one week of delay costs when the asset in question holds your access control. The affected list runs R81 through R82.20, which is a long upgrade tail. Every appliance that has not been refreshed since the last budget cycle is a version decision somebody now has to make.
Check Point's mitigation, if a hotfix cannot land immediately, is to stop the dynamic_server process and prevent it from restarting with dbset process:dynamic_server followed by dbset :save. That disables the Dynamic CLI commands listed in sk144112, so it is a real trade rather than a free fix. Read the SK before you type it.
What to do before Friday
Find the box first. Ask who operates Check Point Security Gateways, Management Servers, Multi-Domain Security Management or Maestro on Gaia R81 through R82.20 in your environment. If an MSP or hosting provider runs your edge, the question goes to them in writing, not to a status page.
Then get Takes and dates. Which Gaia version, which Jumbo Hotfix Take, installed when. For R81.10 or R81, ask for the upgrade plan and whether the dynamic_server mitigation is in place while it gets scheduled. Prioritise anything with the Mobile Access portal reachable from the internet.
Verify the fix rather than the calendar. On a patched appliance, the socket listing for /tmp/dynamic_socket should show owner root, group config, and mode 0660. Before the fix that same entry reads mode 0777, world-writable and owned by root. Check Point puts the exact command in sk1000362. If your appliance still reports 0777, you are exposed regardless of what the change log says.
Then read the logs. The daemon writes every received command to /var/log/dynamicCli.log, and Check Point notes that any command not using the expected Clish wrapper may indicate root execution through the socket. On patched systems, a rejected connection from an unauthorized uid means the escalation was blocked, which is useful confirmation.
Hunt for the rest of it. Published indicators include the scanning addresses 130.94.19.204 and 130.94.19.168, PHP session files containing the string upload_progress_pwn, a process or binary using the CPconfigdaemon naming pattern, tooling under /tmp or /temp, a Dream Annexation note with no payment demand, and unexplained gaps in firewall traffic logs. Check Point's own caveat applies: absence of a match does not rule out compromise, and a match warrants immediate investigation.
One more thing, and it is the part people skip. Patching does not undo a compromise that already happened. A gateway that was exposed and unpatched between 30 October and 4 November is an incident until you have looked, not a maintenance ticket you can close.
The pattern to carry into next quarter
This was not a zero-day race. It was a chained pair where one half had a fix for a week and the other half was a permission bit on a local socket. The remote flaw made the news. The permission bit gave away the machine.
Start with the box nobody admits to owning. Edge appliances outlive the people who installed them, appear on no user-facing inventory, and are usually the last thing anyone volunteers to reboot. That is precisely why they are worth an hour of somebody's Monday.
Sources
- Check Point, sk1000362 - CVE-2026-123039 Gaia DynamicCLI local privilege escalation, affected versions, mitigation and indicators, last modified 12 November 2026: https://support.checkpoint.com/results/sk/sk1000362
- Check Point Research, PHP2Shell Exploitation on Check Point Gateway, 12 November 2026: https://research.checkpoint.com/
- CVE.org record for CVE-2026-123039, published 12 November 2026: https://www.cve.org/CVERecord?id=CVE-2026-123039
- CVE.org record for CVE-2026-109618 (PHP2Shell), disclosed 21 October 2026: https://www.cve.org/CVERecord?id=CVE-2026-109618
- CISA Known Exploited Vulnerabilities catalog, CVE-2026-123039 added 13 November 2026, due 16 November 2026, catalog version 2026.11.13: https://www.cisa.gov/known-exploited-vulnerabilities-catalog
- Rescana, Check Point Gaia CVE-2026-123039 analysis and supplier questions, 14 November 2026: https://www.rescana.com/post/checkpoint-gaia-cve-2026-123039-kev-patch-now
