Back to all articles

PHP2Shell: how compromised Check Point firewalls led to destructive attacks on ESXi hosts

Share

A compromised firewall can become the starting point for an attack on the very infrastructure it is supposed to protect. On 12 November 2026, Check Point Research documented an intrusion chain beginning with PHP2Shell vulnerabilities in an unpatched Mobile Access portal, progressing to local privilege escalation to root through CVE-2026-123039, the installation of multiple backdoors and, in the investigated case, destructive data wiping on a VMware ESXi host.

This is a substantial new development, not a repeat of the original PHP warning. Our October article covering PHP2Shell and its patches examined the two critical vulnerabilities, CVE-2026-109618 and CVE-2026-109619. Check Point's 12 November technical investigation now demonstrates what an attacker was able to achieve after gaining initial access.

Key takeaway: operators of affected Check Point gateways need to verify both PHP2Shell and Gaia DynamicCLI patches, prioritizing internet-exposed Mobile Access portals. If an appliance was reachable while vulnerable during the campaign, patching alone does not establish that it is clean. A targeted search for indicators, persistence and lateral movement may be warranted.

Three CVEs, but two separate stages of compromise

Vulnerability Role in the chain Conditions and limits
CVE-2026-109618 / CVE-2026-109619 — PHP2Shell Remote initial access through a vulnerable Mobile Access web interface Scope depends on the affected product, version, exposed interface and installed hotfix. Check Point lists a 9.2 severity score for the PHP2Shell issues.
CVE-2026-123039 — Gaia DynamicCLI Local privilege escalation to root, after an attacker already has access to the system DynamicCLI executes privileged operations through a local socket with inadequate access control. CVSS 7.8.

CVE-2026-123039 is not, by itself, an unauthenticated remote-code-execution vulnerability. The CISA KEV description says the Gaia DynamicCLI daemon runs as root and listens on a local socket with overly permissive access, without sufficiently authenticating the connecting process.

In the observed attacks, PHP2Shell first provided the execution context from which this local weakness could be reached. The sequence illustrates two separate trust boundaries failing together: an externally reachable application surface and a privileged local control interface.

According to the official Check Point advisory of 12 November, security fixes for affected Check Point deployments had been available since 23 October. The vendor now recommends installing the November Jumbo Hotfix on management servers and security gateways.

From Mobile Access to ESXi: the documented intrusion chain

Researchers describe two waves. The first, on 30 October, targeted Israeli organizations and included destructive activity in some investigated compromises. A broader second wave, on 2 November, scanned and attempted to exploit targets across multiple countries and sectors, with no destructive activity observed in that second wave. The advisory reports observed exploitation through 4 November.

1. PHP2Shell opens the exposed portal

On unpatched systems, the attacker exploited the Mobile Access portal. Investigators found PHP session files containing upload_progress_pwn, alongside numerous Apache/PHP process crashes consistent with repeated exploitation attempts.

These observations must be correlated. A PHP crash alone does not prove successful compromise, nor does a vulnerable PHP package on an unrelated host establish that the Check Point intrusion chain occurred there.

2. Gaia DynamicCLI provides root privileges

The attacker then used CVE-2026-123039 to move from the portal's service-level access to root privileges on the appliance. This allowed the adversary to modify files and services that should have been protected from the less-privileged portal process.

3. Webshells give way to deeper persistence

The first foothold involved PHP webshells placed in template files used by the portal. Researchers describe filenames matching zz.tpl*, as well as a challenge_render.tpl artifact and a temporary upload path, /tmp/.up.tpl.

A second layer was a persistent implant called TSGA, disguised as Check Point-related software. Relevant names include CPconfigdaemon and slot_CPconfigdaemon. TSGA carried a networking component, TSProxy, which was written to /lib/libcpfwcommon.so and could intercept selected traffic while leaving ordinary appliance behavior apparently unaffected.

Check Point describes encrypted remote access, persistence, configuration theft and tampering with binary firewall logs. That last capability materially affects an investigation: a quiet or incomplete firewall log is not reliable proof that no malicious activity occurred, particularly when logging processes and data may have been manipulated.

4. A gateway foothold enables movement toward the hypervisor

In the investigated case, the attacker used a network discovery tool named gogo, then moved from the gateway to a VMware ESXi host over SSH, authenticating with the vpxuser service account.

This does not imply that any compromised Check Point firewall automatically grants access to ESXi. The next step depended on the victim environment's network reachability, service-account credentials and permissions. The critical design lesson is the danger of allowing an exposed security appliance to hold or reach privileged infrastructure identities.

5. A wiper destroys files rather than demanding ransom

On the ESXi host, the adversary deployed destroy_v6, a Linux file-wiping tool. Check Point describes recursive file deletion, with optional overwriting of target data before removal. The malware also left DREAM_ANNEXATION.txt notes containing a political message without a ransom-payment negotiation.

The observed activity should therefore be described as destructive wiping, not as a conventional ransomware encryption campaign. Public reporting does not establish the exact number of ESXi hosts destroyed, and it would be incorrect to assume every scanned or exploited gateway reached this stage.

The timeline: patches existed before the attacks

Date Documented event
21 October PHP upstream fixes for PHP2Shell became available, as reported in our first article.
23 October Check Point says security fixes were available for its affected products.
30 October Initial wave against Israeli organizations; destructive activity in some cases.
2 November A broader wave of reconnaissance and attempted exploitation across countries and sectors.
12 November Publication of Check Point's full investigation and Gaia DynamicCLI advisory.
13 November France's CERTFR-2026-AVI-1529 and the addition of CVE-2026-123039 to CISA KEV.
16 November CISA-listed federal remediation deadline for relevant US government agencies under its directives.

The CISA date is not a French legal deadline. The addition to KEV nevertheless matters operationally because it is evidence of exploitation in the wild, not just a theoretical vulnerability.

To resolve exposure, administrators need to consult both Check Point bulletins: sk1000302 for PHP2Shell and sk1000362 for CVE-2026-123039. The French CERT advisory gives, for the affected product families it covers, the thresholds R81.20 Jumbo Hotfix Take 174 and R82.20 Jumbo Hotfix Take 14. Other branches and platforms require confirmation against Check Point's applicable matrix. The major Gaia release number alone does not prove a corrective hotfix was installed.

Forensic triage: what to investigate on gateways and ESXi

Any investigation should be authorized, documented and proportionate. Preserve relevant evidence where feasible before rebuilding an appliance, while prioritizing containment if destructive activity may still be ongoing.

Evidence source Published leads to investigate Interpretation limits
Mobile Access PHP sessions upload_progress_pwn; session artifacts under /opt/CPcvpn-R81.20/tmp/ in the analyzed environment Paths vary by version; a marker alone does not prove persistent access.
Portal template directory zz.tpl*, challenge_render.tpl, /tmp/.up.tpl Compare with trusted releases; preserve metadata, timestamps and hashes.
Gaia processes and persistence CPconfigdaemon, slot_CPconfigdaemon, /tmp/.bd.log Host-side visibility can be incomplete on a privileged compromised appliance.
Libraries and communication /lib/libcpfwcommon.so, Unix socket /tmp/.p_sock, unusual outbound connections Binary firewall logs may have been altered by the implant.
Campaign network origins 130.94.19[.]204, 130.94.19[.]168 An IP match is a lead, not proof of successful exploitation; no match does not clear a system.
ESXi and backup systems Unexpected SSH logins, use of vpxuser, destroy_v6, DREAM_ANNEXATION.txt Confirm any loss or deletion against host, hypervisor and storage evidence.

These artifacts come from Check Point Research's technical report and its vendor security advisory. The original report includes SHA-256 hashes for several recovered samples. Hashes should be used to corroborate findings, not as the only test for compromise.

Correlate independent evidence: remote-access logs, network telemetry, centrally collected events, host monitoring, hypervisor audit records, service-account activity and backup-system access. Because TSGA can manipulate appliance logging, evidence stored outside the compromised gateway becomes especially valuable.

Patching stops an exploit path; incident response must remove the foothold

A patch closes a vulnerability but may not remove an installed webshell, a persistent backdoor, stolen credentials or access added during the breach. The response plan must address the full attack chain.

  1. Contain exposure: identify internet-facing Mobile Access portals, apply both relevant hotfix families and restrict management interfaces to appropriate trusted networks.
  2. Reconstruct the vulnerable window: establish when each exposed appliance was vulnerable and when its fixed build and hotfix take actually entered production.
  3. Hunt for persistence and lateral movement: investigate webshells, TSGA/TSProxy, unexpected account or configuration changes, suspicious network connections and SSH access to the virtualization layer.
  4. Revoke potentially compromised credentials: rotate passwords, API tokens, SSH keys and service-account secrets that were accessible to the gateway or used for the attack, based on actual exposure.
  5. Restore trust: where privileged compromise is established, use a trusted rebuild or recovery process rather than assuming that deleting a few suspicious files is sufficient.
  6. Validate independent backups: verify backups do not rely on the same management credentials, network paths or hypervisor storage available to the intruder.

A VM snapshot is not automatically an independent backup. If the attacker can reach the datastore, the backup orchestrator or the identity controlling them, snapshots may also be lost. Separate access rights, immutable or offline copies, and tested restoration procedures reduce the impact of such an intrusion.

This is the distinction between fixing a software defect and restoring a compromised infrastructure to a demonstrably trustworthy state.

Attribution: shared tooling is not proof of who ran the campaign

Check Point identified significant similarities between TSGA/TSProxy and PhiliKit, previously associated with intrusions into Ivanti devices. ESET had linked the earlier PhiliKit activity to a China-nexus actor, UNC5221.

However, the Check Point researchers explicitly state that shared code and cryptographic material do not provide sufficient confidence to attribute this operation. Direct involvement, access brokerage and deliberate false attribution are among the scenarios they discuss.

The political message left in DREAM_ANNEXATION.txt does not establish the operators' identity or objective either. The motive and attribution therefore remain unresolved.

The enterprise lesson: the management plane is part of the blast radius

Firewalls, VPN portals, management consoles and hypervisors are part of the infrastructure control plane. Compromise at one perimeter device can cross the trust boundaries intended to protect workloads, data and backups.

For GLOV Consulting, the case highlights three governance priorities: maintain an accurate appliance-and-hotfix inventory, segment privileged infrastructure access, and test a recovery strategy that does not depend entirely on the attacked environment.

Our earlier coverage of PHP2Shell and its PHP fixes and Node2Shell in Elastic Synthetics explores other ways that a single vulnerable software component can threaten a broader deployment. This article focuses on a documented intrusion chain and destructive outcome, not a general claim about all installations.

GLOV's specialized digital-forensics service is still marked as upcoming on the website. This is an educational technical analysis, not an announcement that an upcoming service is commercially available.

Information reviewed as of 14 November 2026. Public reports do not conclusively establish the overall victim count, the full destruction scope beyond the investigated incidents, or the identity of the operators.

Official sources and independent analyses

Related articles

Back to all articles