ZERO-DAY EXPLOITATION · CVE-2026-93616

Check Point Management Server Zero-Day

61 Days From Exploit to Fix. Attackers ran unauthenticated code on Check Point Security Management Servers on July 23 using CVE-2026-93616, a 9.8 path traversal bug that Check Point only patched on September 22.

Matt Lucas  |  September 22, 2026  |  5 min
Editorial hero illustration
9.8
61 days
July 23 exploit to September 22 fix
3
attacker certificate subjects named
11
affected management server releases
TL;DR
  • What: Check Point patched CVE-2026-93616 on September 22, a path traversal in the Security Management Server web service that attackers used for unauthenticated script execution in targeted attacks on July 23.
  • Impact: The management server holds firewall policy for every gateway it manages, so code execution there is control of the ruleset, not a single host compromise.
  • Fix / mitigation: Install the fixed Jumbo Hotfix take listed in Check Point support article sk1000171, and note that LivePatch Take 28 or 29 from September 16 does not remediate this CVE.
  • Who's at risk: Anyone running Security Management on R82.20, R82.10 Take 44 or below, R82 Take 126 or below, R81.20 Take 166 or below, or any end of support release, plus Spark and Security Gateway owners exposed to CVE-2026-85102.

Check Point patched a zero-day in its Security Management Server on September 22, 61 days after attackers used it. The company says a handful of targeted attacks on July 23 exploited CVE-2026-93616, a path traversal bug in the management server web service that lets an unauthenticated attacker upload and run scripts. Check Point rated it 9.8 out of 10 on CVSS. The management server is the box that holds firewall policy for every gateway it manages, so code execution there is not a single host compromise, it is control of the ruleset.

A path traversal that ends in script execution

The web service on the management server does not properly restrict which files and folders a request can reach. An attacker who can reach that service writes a script to a location the server will run, then triggers it. No credentials, no session, no prior foothold beyond network access to the web service.

Check Point's advisory does not say what network position an attacker needs, and that gap matters. A management server reachable only from an internal admin VLAN is a very different risk than one answering on a public address. The advisory also does not name the July targets, does not attribute the activity, and does not describe what the attackers did after exploitation. Hunting guidance and indicators of compromise are in support article sk1000171. Installing the fix tells you nothing about whether you were hit before it, so treat patching and hunting as two separate tasks with two separate owners.

LivePatch Take 28 does not cover this

On September 16 Check Point fixed a different management server flaw, CVE-2026-91843, through LivePatch Take 28, or Take 29 on R82.20. Check Point says those takes do not remediate CVE-2026-93616. If your change record reads "LivePatch applied September 16" and the ticket is closed, the server is still exposed.

The version math catches people who already patched

Check Point numbers Jumbo Hotfix updates for each release by Take, and the LivePatch channel uses a separate take sequence. The CVE record lists these management server versions as affected:

Here is the trap. CVE-2026-85103, a VPN certificate flaw Check Point fixed on September 9, affected management servers as well as gateways. On R82.10, R82 and R81.20 the affected range for CVE-2026-93616 runs one take higher than the range for that September flaw. A server patched just far enough to clear the September 9 issue is still vulnerable to the September 22 one. Note also the disagreement in the source material: the CVE record qualifies R82.20 as affected only with no Jumbo Hotfix installed, while Check Point's own advisory lists R82.20 with no condition. Assume the broader reading until Check Point clarifies.

Second front: Spark firewalls under active probing

Since September 12, attackers have been trying to exploit CVE-2026-85102, a flaw in how Check Point gateways validate certificates while a VPN connection is being established. It may let an unauthenticated attacker run code on the gateway. Check Point shipped the fix on September 9 with no evidence of exploitation at the time. Three days later the attempts began, concentrated on Spark, the small business firewall line. That is the familiar sequence: the patch is the disclosure, and the window between release and weaponization is now measured in days.

Affected products are Security Gateway and Spark, centrally or locally managed, on R81 and R81.10 (both end of support), R81.10.x, R81.20, R82, R82.00.x and R82.10. The Netherlands NCSC notes the flaw applies when these products use Site-to-Site VPN or Remote Access VPN. Check Point says the attempts arrived from anonymizing infrastructure including commercial VPN services and proxies, using certificates with these subjects:

That list is explicitly incomplete. Do not build a detection that only matches those three strings. Hunt on behavior instead: any certificate based Mobile Access login you cannot tie to a known user or a known source, followed by internal port and service scanning from the gateway's client pool. Check Point says customers who installed the September 9 fix are protected, but the advisory does not say whether any attempt succeeded.

If a gateway cannot be patched this week

For Site-to-Site VPN, the NCSC lists a Check Point workaround: turn off the implied VPN rules and permit UDP 500 and 4500 only from specific peer IP addresses. The workaround does not apply to locally managed Spark firewalls, which need the fix in sk1000117. Remote Access VPN gets no equivalent workaround, so those gateways are a patch or accept the risk decision.

What to do this week

cpinfo -y all 2>/dev/null | grep -iE 'jumbo|take|R8[0-2]'
clish -c "show version all"
clish -c "show installer packages installed"
fw log -n -p $FWDIR/log/fw.log | grep -iE 'CN=vpn,|CN=vpn-user,|CN=vpnuser,'

Fix It Yourself

Everything above is what happened. This is what to run. The check commands change nothing.

Check whether you are exposed

Print the installed release and Jumbo Hotfix take on a Check Point management server or gateway, changes nothing

cpinfo -y all 2>/dev/null | grep -iE 'jumbo|take|R8[0-2]'

Show the Gaia version and the CPUSE packages already installed, read only

clish -c "show version all"; clish -c "show installer packages installed"

Search current VPN and Mobile Access logs for the certificate subjects Check Point published for CVE-2026-85102

fw log -n -p $FWDIR/log/fw.log | grep -iE 'CN=vpn,|CN=vpn-user,|CN=vpnuser,'

Close it

Take a Gaia snapshot before touching the box so the hotfix is reversible and pre-patch state is preserved for forensics

clish -c "add snapshot name pre-cve-2026-93616-$(date +%Y%m%d)"

List the available CPUSE packages, then download and install the fixed build named in sk1000171 by substituting its package number

clish -c "show installer packages available-for-download"
# substitute the package number for the sk1000171 build below
clish -c "installer download <PACKAGE_NUMBER>"
clish -c "installer install <PACKAGE_NUMBER>"

No workaround exists for Remote Access VPN on CVE-2026-85102, so confirm which VPN listeners are actually exposed before deciding to accept risk or patch out of cycle

netstat -an | grep -E ':500 |:4500 |:443 '

Prompts for the agent you already run

Paste this when you need to know which Check Point management servers in your fleet are exposed to CVE-2026-93616

You are auditing Check Point Security Management Servers for CVE-2026-93616. For each host in my inventory file, run only read-only commands: 'cpinfo -y all' and 'clish -c "show version all"'. Parse the release (R82.20, R82.10, R82, R81.20, R81.10, R81, R80.x) and the Jumbo Hotfix Take number. Flag a host as VULNERABLE if it is R82.20, or R82.10 at Take 44 or below, or R82 at Take 126 or below, or R81.20 at Take 166 or below, or R81.10 at Take 190 or below, or any R81 or R80.x release. Print the raw command output you used for each determination next to the verdict so I can check your parsing. Note explicitly that LivePatch Take 28 or 29 from September 16 does NOT remediate this CVE, so do not treat it as a fix. Produce a table of host, release, take, verdict. Do not install anything, do not restart any service, and stop after printing the table.

Paste this when hunting for CVE-2026-85102 exploitation attempts against gateways and Spark firewalls

Hunt for exploitation attempts of Check Point CVE-2026-85102 in my VPN and Mobile Access logs. Search from September 10 2026 forward. First, find every certificate based Mobile Access or VPN authentication event and extract the certificate subject or DN field, the source IP, the username, and the timestamp. Flag events whose subject matches CN=vpn, CN=vpn-user, or CN=vpnuser with OU=users and O=global, but also flag any certificate subject that does not map to a user in our directory, because Check Point says the published list is incomplete. Second, for every flagged session, pull the subsequent connection records from that assigned IP and highlight any session touching more than 20 distinct internal destination ports or hosts within 10 minutes, which is the internal scanning behavior Check Point described. Show me the exact log queries you ran and the raw matching lines, not just counts. Do not delete, rotate, or modify any log file, and stop before making any firewall policy change.

What to alert on

The single highest value observable for the gateway side is the certificate subject field on Check Point Mobile Access and VPN authentication events, visible in SmartView or in exported fw.log as the certificate DN alongside the user and source fields. Alert on any certificate based Mobile Access login where the subject does not resolve to a provisioned user in your directory, and treat the published subjects CN=vpn, CN=vpn-user and CN=vpnuser with OU=users,O=global as confirmed bad rather than as the whole list. Pair that with a second rule on the session that follows: an authenticated VPN client IP touching more than roughly 20 distinct internal destinations or ports inside 10 minutes is the internal reconnaissance Check Point observed, and it is the part attackers cannot easily hide by rotating certificate subjects. On the management server side, where no exploitation signature has been published, watch process lineage on the host: the web service process spawning a shell, an interpreter, or any child it does not normally fork, plus creation of new script files under the web accessible directories. Those two events, plus any inbound connection to the management web service from outside your admin subnet, are what to alert on today while you work through sk1000171.

Questions about your exposure?

RedEye Security provides assessments for organizations that need to understand their real risk.

Talk to us