November 13, 2026
Check Point Scalable Platforms Privilege Escalation via RCE and Arbitrary File Write
On Check Point R82 security gateways with ElasticXL enabled and Maestro SGMs (all versions), an attacker with Access Policy / NAT write priv

By Konstantin Burov
5 min read
Check Point Scalable Platforms Privilege Escalation and Arbitrary File Write by an Authenticated Administrator
On Check Point R82 security gateways with ElasticXL enabled and Maestro SGMs (all versions), an attacker with Access Policy / NAT write privileges or access to an isolated network for cluster synchronization can escalate their privileges and execute OS commands and write to an arbitrary file with superuser rights.
Discovery
I first encountered Check Point Maestro several years ago. At some point, I realized that the mechanism for executing global commands to manage cluster members (g_fw, g_tcpdump, g_all, etc.) did not use SSH or any other functionality such as cprid with SIC protection and that's rather interestingβ¦
Traffic capture revealed that commands are transmitted directly between nodes via the synchronization (Sync) interface, using TCP port 1130 within the 192.0.2.0/24 network without any authentication or encryption.
You might ask: "It's an isolated network, and all nodes connect directly to the orchestrator β why even consider this a vulnerability?" And you'd be right, but as always, there's a catchβ¦
Unfortunately, I didn't have a Maestro of my own on hand to test the concept, so I shelved the idea until a better time.
ElasticXL
With the release of R82, a 'cheap' version of Maestro has arrived, giving me the chance to explore this new offering. ElasticXL makes it possible to build a cluster similar to Maestro though without an orchestrator and limited to 3 members per Site and, most importantly, it can be deployed on a virtual machine.
In R82, some commands from expert mode have been migrated to the gClish shell. The cluster operation logic and the process for adding members to ElasticXL have changed; however, the synchronization network (192.0.2.0/24) on the Sync interface remains in place, as do the global commands.
Port 1130 is listened by the gexecd daemon (/usr/libexec/gexecd) at address 192.0.2.* on the Sync interface.
There is a client utility, /usr/bin/gexec, used for sending commands. Global commands all make use of it in one way or another.
Here is a small example of how to run it:
Exploitation
Since the service listens on a port only via the Sync interface, you need to connect to that network. Fortunately, doing so between VMs is very straightforward.
Since the service operates without encryption and the protocol itself is very simple, a one-liner like this will send a command for execution and output the result:
echo -en β\xf0\x0c\x00\x00\x00<ver>2</ver>\x0a\x00\x00\x00<vs>1</vs>\x2b\x00\x00\x00<req><mode>finite</mode><cmd>id</cmd></req>β | nc 192.0.2.1 1130 | grep -oaP β</vs><outp>\K(.*)βecho -en β\xf0\x0c\x00\x00\x00<ver>2</ver>\x0a\x00\x00\x00<vs>1</vs>\x2b\x00\x00\x00<req><mode>finite</mode><cmd>id</cmd></req>β | nc 192.0.2.1 1130 | grep -oaP β</vs><outp>\K(.*)β
And if anyone other than you has access to the ElasticXL cluster synchronization network (which is hopefully on a separate VLAN) that person could potentially gain privileged access to your security gateways.
And what about Maestro β given that the synchronization network there is isolated and connected directly to the orchestrator (Admittedly, there's a question regarding the dual site synchronization network, but I have no way to verify that)?
Since this is standard TCP, in purely theoretical terms, any host from the inside could connect to the service, provided the traffic passes through the gateway. This would, of course, require the administrator to have decided to allow everything for everyone β which is unlikely, though not impossible.
To address this, CheckPoint has implemented a safeguard: access to the Sync interface is blocked at the Firewall VM kernel level, and packets are dropped with the reason 'Packet Source or Destination is Sync'. It's unfortunate, but since we know that Access policies are applied before NAT, perhaps I can bypass this simple restriction β let's see what happens.
Access Log:
The logs show an 'Accept' event, indicating that a connection is accepted on an external IP and then forwarded to the address of the 'Sync' interface. This means that an administrator with access to Access Policy and NAT settings (not necessarily a 'SuperUser') could create a 'backdoor 'to execute commands on the gateway with the privileges of the 'admin' (root) user. The Source or Destination addresses do not matter (except this does not belong to the Sync network); what matters is that the traffic passes through the gateway and the NAT rule is triggered correctly.
I wrote a script in Python as a PoC, the link will be below.
Arbitrary File Write
In addition to gexec, there is a service responsible for copying files between cluster nodes - gcopyd, which runs on port 1132. This service is used by the asg_cp2blades command. The issues and the logic behind its operation are exactly the same as those of the aforementioned service.
Daemon β /usr/libexec/gcopyd
Client β /usr/bin/gcopy
gcopy PoC
It's a somewhat rough diagram, but I hope it's clear:
Access rule for gexec and gcopy:
NAT rule for gexec and gcopy:
FIX
For the NAT scenario fix is ββalready available starting with releases R82.10 and R82.20; for R81.20 and R82, the latest JHFs must be installed. The firewall kernel now blocks access to the Sync interface, even when NAT is involved (sk1000364).
Regarding the direct connection to the synchronization interface on ElasticXL, Check Point has added MACsec data encryption (R82.20), which must be explicitly enabled in the configuration. I suspect this introduces additional overhead during cluster synchronization.
Now I understand why the fix took so long: resolving the issue involving NAT-based deception is not difficult, but implementing MACsec is a significantly more complex task β and such changes are generally not included as part of a Jumbo Hotfix.
When capturing traffic, we only see the MACsec encrypted payload:
We can also see a new interface β SecureSync:
MACsec settings can be viewed using standard Linux commands:
Afterword
The g_all set of commands can be traced back as far as version R76.10SP specifically in the upgrade manual for 60000/40000 series appliances, which was released in Q4 2014. It is therefore safe to assume, with a reasonable degree of certainty, that this vulnerability has existed within the Scalable Platform for about 12 years.
CVE-2026β120490 was assignet by Check Point CNA
Links
β PoC GitHub repository β Check Point Maestro Global Operating System Commands β Secure Synchronization for ElasticXL β R82.20 Release Notes β CVE-2026β120490 β SK1000364
Timeline
25β11β2024 β Initial contact 28β11β2024 β Confirmation of vulnerability and patch plans from vendor 25β02β2025 β Patch status request (90 days since initial report) 21β04β2025 β One more tryβ¦ 19β08β2025 β CVE escatation for CNA issue 07β10β2026 β Disclosure coordination with vendor 16β11β2026 β Disclosure