Skip to content
RESEARCH INDEX BREACHROAD / INTELLIGENCE NOTE

Check Point PHP2Shell and CVE-2026-123039 under attack

Check Point reports active attacks on Mobile Access gateways: PHP2Shell provides RCE and CVE-2026-123039 elevates privilege in Gaia. Learn how to respond.

PUBLIC RESEARCH
AUTHOR
/ CEO of Breachroad · OSCP · PNPT
PUBLISHED
13 November 2026
READING TIME
19 min read
TOPIC
Vulnerabilities and CVEs
Check Point PHP2Shell and CVE-2026-123039 under attack

Check Point Research has documented active exploitation of Security Gateways beginning with PHP2Shell flaws in Mobile Access portals and continuing through CVE-2026-123039, a local privilege escalation in Gaia. The vendor has observed scanning and exploitation attempts since 30 October 2026 and released fixes in November Jumbo Hotfix Accumulators.

Primary sources are the technical PHP2Shell exploitation on Check Point Gateway analysis and the active-exploitation security advisory. Verify fixed builds in the release notes for the relevant branch, such as R82.10 Take 52, because hotfix take numbers vary.

The confirmed chain

The first component is PHP2Shell, assigned CVE-2026-109618 and CVE-2026-109619, which can provide remote code execution in the vulnerable portal layer. The second is CVE-2026-123039, a Gaia local elevation issue involving a globally writable DynamicCLI control socket. An operator with initial code execution can use it to gain higher privilege.

Check Point assigns CVSS 7.8 to CVE-2026-123039 and lists Gaia R82.20, R82.10, R82 and R81.20. The local flaw is not itself the remote entry point; its importance increases when chained with preceding RCE. This is why a single CVSS score without an attack graph can understate priority.

Who is exposed

The most direct risk applies to unpatched gateways with an accessible Mobile Access portal. Review every node, role, version and configuration, including standby, test and separately managed instances. A public address may lead through a load balancer or NAT to a system absent from routine scans.

Determine whether the portal is enabled, who uses it, which Gaia branch runs on every node and the installed Jumbo take. A major release number without take is insufficient. An external attack-surface inventory should record function and ownership, not only an IP.

Why chaining changes risk

The first vulnerability provides execution in a constrained context. The second converts local access into higher privilege. An adversary can then attempt configuration changes, persistence, theft of appliance data or movement through the gateway. Each step may fail, but the reachable path is more dangerous than the sum of two scores.

In a threat graph, entry, execution primitive and escalation lead to an objective. Patching, exposure controls, minimal services and monitoring can cut different edges. Address both issues; do not dismiss a local CVE because it requires prior access when that access path is actively used.

Immediate actions

Identify every gateway, release and active portal. Preserve logs and state before restart. Obtain the right Jumbo Hotfix through an authorised source, validate compatibility and install according to vendor procedure. In a cluster, confirm order, synchronisation and the build on each member.

If the update needs a window, restrict the portal to authorised sources or temporarily disable it where supported and operationally possible. This reduces exposure, but public scanning began before complete hotfix rollout, so historical analysis remains necessary.

After updating:

  1. confirm release and take on every node;
  2. test Mobile Access and authentication;
  3. inspect HTTP/PHP logs, processes and file changes;
  4. verify configuration, accounts, keys and scheduled jobs;
  5. correlate traffic from the gateway into internal networks;
  6. rebuild from a clean image when integrity cannot be established.

Gateway hunting

Look for unusual portal requests, PHP errors, new or modified files, processes started by the web layer, outbound connections and attempts to access DynamicCLI mechanisms. Review temporary directories, configuration modifications and administrator events with no matching change request.

Compare state to a trustworthy baseline: packages, file hashes, policies, accounts, SSH keys, certificates, crontab and rules. A baseline created after scanning began may include compromise, so record that limitation.

At the network layer, inspect DNS and appliance egress. A gateway normally talks to many services, so connection count alone is not useful. A new ASN, domain, port or rhythmic beacon correlated with a web event is a stronger signal.

Suspected compromise

A security appliance is unusual evidence: it both protects traffic and may have been altered. Avoid ad-hoc cleanup. Reduce exposure, preserve an image and logs where the platform permits, then prepare a clean node using a verified image and policy.

Rotate secrets accessible to the gateway: directory accounts, keys, certificates, API tokens and management credentials. Close re-entry first or new secrets may be captured. Review the management system and administrator identities because a compromised gateway can provide material for follow-on access.

Failing over to an unpatched standby is not safe recovery. Nodes may share configuration and exposure history. Recovery needs a documented trust point and repeated validation.

Validating the fix

Release and take are initial evidence, not the final proof. In a controlled environment, confirm that the portal rejects invalid input, the web context cannot use the sensitive DynamicCLI interface, and attempts are logged. Do not run a public PoC against production without authorisation and a stop plan.

Also update disaster-recovery nodes, installation images and new-node procedures. If automation builds gateways from an affected snapshot, the issue returns at the next failure. An infrastructure penetration test should cover management plane, portal and post-entry segmentation.

Hardening edge appliances

Expose only required public services. Restrict management to a dedicated network and strong authentication. Separate administrative accounts, export logs to an independent system and alert on policy changes, new keys and unscheduled restart.

Set an SLA for actively exploited edge flaws in hours rather than weeks. Maintain emergency cluster-patching capacity. A vulnerability-management programme should prioritise KEV status, exposure and chaining rather than CVSS alone.

Check Point facts and Breachroad analysis

Source facts: Check Point has observed active exploitation since 30 October; PHP2Shell comprises CVE-2026-109618 and CVE-2026-109619; CVE-2026-123039 is local Gaia privilege escalation through a writable DynamicCLI socket; November hotfixes cover the listed branches. Breachroad analysis: the threat graph, IR sequence, baseline and hardening are operational recommendations. We do not attribute the activity to an actor without a source.

The gateway owner’s plan

Today: verify exposure, preserve data, install the correct Jumbo and begin hunting from 30 October. This week: validate integrity, rotate secrets where warranted, fix DR and improve alerts. This month: rehearse clean recovery and review every edge appliance.

A security audit can determine whether other public gateways carry similar update debt. The central lesson from PHP2Shell and CVE-2026-123039 is that local privilege escalation cannot be assessed in isolation when attackers already have a confirmed route to code execution.

Policy integrity control

After an incident, compare active policy with an approved export and change repository. Find new objects, rules, exceptions, NAT, administrators and logging modifications. A normal-looking GUI is insufficient; high privilege may alter presentation or event recording too.

Every difference needs a change ID or owner confirmation. Automated diff helps with scale, but a human assesses rule order and impact. Valid syntax does not imply safe policy.

Management Server and change propagation

A gateway may be centrally managed, so examine the Management Server, administrator sessions, publications and policy installs. Determine whether a suspicious change was local or pushed centrally. Compromise of either component has a different scope and recovery path.

Separate operating-system login from policy-publication identities. Require MFA and approval for changes spanning gateways. Export audit logs outside the Check Point administration domain so a compromised panel cannot remove the only copy.

VPN certificates and credentials

Mobile Access gateways may hold server keys, trusted CAs, RADIUS secrets and IdP integrations. Under suspected root compromise, map and assess rotation of each. Install replacement certificates only on a fixed, verified appliance. Review revocation lists and client caches too.

Where the gateway brokers authentication, correlate sessions with IdP records. An unexplained session lacking MFA evidence may indicate further trouble. Disabling a user does not necessarily revoke a device token or live VPN session.

Detection that survives IOC changes

Instead of one address, monitor a sequence: unusual portal request, web-layer child process, access to a local privileged interface, file or configuration modification and new egress. Each event may be legitimate alone; together they form a stronger hypothesis.

Test the rule against history and maintenance activity. Document sources and delay. If events exist only locally, configure export before the next incident.

Maintaining Jumbo Hotfix levels

Report branch, take, release date and owner for every gateway monthly. Installation images and standby nodes follow the same lifecycle. For urgent takes, maintain a fast test path covering VPN, routing, cluster and logging.

Track advisory-to-last-node time, exceptions, public portals and oldest searchable log. Technical-team security training should rehearse emergency cluster updates so a critical fix does not wait solely because failover is feared.

Gateway incident-closure criteria

Release and take must be correct on every node, not only active. Resolve policy differences, identities, keys, PHP child processes, egress and DynamicCLI use. A suspicious system requires clean recovery and rotation of reachable secrets.

The report states the beginning of the examined period and retention gaps. No IOC is not proven integrity. The service owner accepts residual risk and a 30-day verification plan.

After one month, verify that the cluster did not return to an old take, portal exposure remains correct and behavioural rules create the expected alert. Rehearse in test rather than public production. This proves fix durability without risky reproduction of a full exploit.

Attach the result to the edge-appliance register and reuse it as the baseline for the next Jumbo update.

SHARE / COPY