< ciso
brief />
vCenter Zero-Day, Critical PHP and NGINX Patches Lead the Day

vCenter Zero-Day, Critical PHP and NGINX Patches Lead the Day

Coverage: 19 Nov 2026 (UTC)

< view all daily briefs >

Critical fixes for core infrastructure software dominate November 19, 2026. Broadcom patched a VMware vCenter authentication bypass that, according to Google, a China-linked group exploited as a zero-day, while the PHP and NGINX projects shipped updates for multiple critical remote code execution flaws. Exploitation attempts against the Java4Shell OpenJDK bug, new CISA KEV entries, a compromised Microsoft package on PyPI and a confirmed intrusion at the European Commission complete the day.

VMware vCenter Bypass Exploited Before Patch

Broadcom published an advisory for vCenter describing CVE-2026-123890, a critical authentication bypass in the VMware Directory Service rated 9.8 on the CVSS scale. An attacker with network access can bypass authentication and gain unauthorized access, and no workaround is listed. The flaw affects vCenter 8.0, 9.0 and 9.1, along with builds bundled in VMware Cloud Foundation (including the 5.x line), vSphere Foundation and the Telco Cloud products; older releases such as vSphere 6.5, 6.7 and 7.0 should be presumed vulnerable. Fixed vCenter versions are 9.1.1.0200, 9.0.2.0200 and 8.0 U3m, VMware Cloud Foundation 5.x is fixed in 5.2.5, and vCenter 7.0 customers receive a patch only with an extended support contract. Broadcom classifies the issue as an emergency change under ITIL methodologies, notes that the updates are cumulative and cannot be applied through vCenter Quick Patch, and warns that installing 8.0 U3m or 9.0.2.0200 blocks a later upgrade to VCF 9.1.1, so upgrade paths require careful planning.

The advisory also fixes four lower-severity bugs in SDDC Manager and VMware Tools for Windows, including an SDDC Manager SQL injection (CVE-2026-123891, 8.1) and a VMware Tools SAML token signature bypass (CVE-2026-123894, 6.7), all requiring authenticated or privileged access. This is the second critical Directory Service flaw this year: a July traversal bug, CVE-2026-59310, was exploited within five days and tied to Babuk-derived ransomware on ESXi hosts, compromising roughly 361 IP addresses across 47 countries. Broadcom's advisory does not mention in-the-wild exploitation of the new bypass, which was reported by Mandiant.

Mandiant and Google Threat Intelligence Group have separately detailed ARMORY CASTLE exploitation of the same flaw, which, according to Google, lasted more than two months before a patch was available. The suspected PRC-nexus actor, formerly tracked as UNC5221 and also known as Warp Panda, first gained access through a BeyondTrust Remote Support zero-day, CVE-2026-102764, then moved laterally to internal Linux appliances and scanned for ports 9080, 5480 and 2012, the last being the vCenter SSO control interface. In early September 2026 it exploited CVE-2026-123890, described as an unauthenticated SRP-6a session key disclosure over DCE/RPC that yields an authentication bypass as administrator@vsphere.local. The exploit was delivered through BONESAW, a Golang tool that exfiltrates the LMDB database, extracts the STS SAML signing key, forges administrative tokens and creates an SSO admin account. The actor also deployed the SLIMSAVAGE and CAGEMATCH webshells, the BEARBONES and LAZYBONES backdoors and the SAWSNAKE SAML utility, escalated to root by setting SUID on /bin/env through gdb, took the vCenter master encryption key and database credentials, and then moved to an ESXi host.

A BONESAW build timestamp of July 13, 2026 suggests the group knew of the flaw before the related CVE-2026-59309 was disclosed on July 29, and the tool was used against a server already patched for that bug. BONESAW works against vCenter 7.x and 8.x, yet vCenter 7.0, which reached end of general support in October 2025, has no public patch. Mandiant CTO Charles Carmakal urges organizations to both patch and examine their systems for compromise.

PHP and NGINX Close Critical Web Stack Flaws

The PHP project shipped a PHP security release on November 19, 2026, fixing seven flaws rated critical across the supported 8.2 to 8.5 branches. The maintainers had warned a day earlier that the critical issues require non-default configurations or uncommon setups. The flagship bug, phpAfterShell (CVE-2026-126714), was reported by Depthfirst and affects the default file-based session handler: an interrupted session write during upload progress can leave a malformed file that session_start() later deserializes, leading to code execution when paired with a gadget chain in the application. A separate PHP-FPM request smuggling flaw, CVE-2026-129289, rated 9.2, can allow unauthenticated remote PHP code execution in nginx, Apache, HAProxy and LSAPI configurations that reuse backend connections to PHP-FPM. Other fixes cover use-after-free bugs, a PHP-FPM local privilege escalation and a PHAR OpenSSL signature bypass.

Some of the new fixes complete work left open by October's PHP2Shell patches for CVE-2026-109618 and CVE-2026-109619. CVE-2026-109618 is listed in the CISA Known Exploited Vulnerabilities catalog, and the PHP2Shell flaws have seen exploitation against exposed Palo Alto Networks management interfaces and Check Point Security Gateways. Servers patched for PHP2Shell on October 21 remain vulnerable to phpAfterShell. Fixed versions are 8.5.13, 8.4.28, 8.3.37 and 8.2.36, with PHP 8.6.0 also carrying the fix; PHP 8.1 and earlier are end-of-life, though Ondrej Surý and Aikido have released patched packages for unsupported versions dating back to PHP 5.6. Interim mitigations include disabling session upload progress with session.upload_progress.enabled=0 and turning off FastCGI connection reuse, although these do not cover every flaw. Wiz reports no observed exploitation of the new bugs. The highest-risk deployments are PHP-FPM behind connection-reusing front ends, shared hosting and products that bundle their own PHP runtime, and teams are advised to rebuild container images, restart PHP workers and hunt for unexpected shells, new PHP files and unusual outbound connections.

Researchers disclosed NGINX2Root on November 18, 2026, a set of five critical nginx vulnerabilities that allow an unauthenticated remote attacker to execute code, and F5 and the nginx project released patches the same day. The two most severe, CVE-2026-123007 and CVE-2026-123633, each carry a CVSS 4.0 score of 9.2. The first is an HTTP/2 flow-control integer underflow, present since version 1.9.5, that corrupts worker memory; the second is a use-after-free affecting nginx when it proxies HTTP/2 requests to a backend. The remaining flaws are use-after-free bugs in the DNS resolver, QUIC/HTTP/3 and the mail proxy. Xint researchers reached root by chaining CVE-2026-123007 with an unpatched Linux kernel SCTP flaw to escape a Docker container, while Wiz demonstrated remote code execution for both headline bugs.

Rapid7 published a proof-of-concept for CVE-2026-123007, but it requires roughly 128GB of HTTP/2 traffic, an x86-64 glibc build and an ASLR bypass, which limits mass exploitation while still allowing targeted attacks. No in-the-wild exploitation has been confirmed, none of the flaws were in the CISA KEV catalog as of November 19, and CERTs in Japan, Poland and France issued alerts. Censys estimated about 8.8 million services potentially exposed, with over 511,000 already patched or mitigated, while researcher Florian Roth cautioned that exploitation requires non-default configurations and that version-based exposure claims can mislead. Fixed releases are nginx Open Source 1.31.9 and 1.30.7 and NGINX Plus 37.3.0.1 and 37.0.8.1, and the service must be fully restarted rather than reloaded. Disabling HTTP/2 mitigates only the two headline CVEs and affects gRPC services that depend on it, and the retired Kubernetes ingress-nginx project will receive no fix.

Java4Shell Probing, KEV Additions and ASUS Fixes

Threat actors have begun targeting Java4Shell, tracked as CVE-2026-114938, a memory corruption bug in how OpenJDK 17.0.6 and newer handle Base64 data. Exploitation requires an x86-64 processor with AVX-512 VBMI support, and ARM-based systems are unaffected. Calif disclosed the flaw on November 17, the same day Oracle and the OpenJDK project released patches, and reported unauthenticated remote code execution against default setups of Jira, Confluence, Jenkins and Keycloak, each requiring a tailored attack chain. GreyNoise, WatchTowr, Cloudflare and Previdian have observed scanning and blind exploitation attempts, with one IP linked to both Java4Shell and the AI pentesting framework Artex. No successful compromises have been confirmed, though WatchTowr expects product-specific exploits to mature in the coming days. Vendor responses vary: Cisco has confirmed more than a dozen affected products with no fixes or workarounds yet, Siemens advises taking affected Industrial Edge Management instances offline, Elastic lists Elasticsearch 7.17.3 onward and Logstash 8.6.2 and later as affected, Red Hat rates the issue Critical across RHEL 8, 9 and 10, Atlassian suggests disabling AVX-512 for Data Center products, Jenkins urges updating Java and restarting, and Amazon, Microsoft and Azul have released patched OpenJDK builds.

CISA added four exploited flaws to its Known Exploited Vulnerabilities catalog, each permitting remote code execution. CVE-2026-90436 in Apache Axis was exploited with the 4EverAble2Shell bugs to deploy cryptocurrency miners, and Rapid7 observed it used against Atlassian Confluence Data Center, where attackers hunted for credentials and payment card numbers and signed in to Microsoft 365 within about 12 hours. Axis 1.x is end-of-life and unpatched, so users should migrate to Apache Axis 2/Java or apply a July 12, 2026 commit. CVE-2026-90817 in REDCap was abused to stage a web shell and deploy XMRig disguised as syslog-ng, with persistence through cron and an SSH key. CVE-2026-40217 in BerriAI LiteLLM was used in LLMjacking attacks to escalate limited inference keys into full administrative credentials, according to Google Threat Intelligence Group, and CVE-2024-10441 in Synology DiskStation Manager was exploited in NAS attacks that deployed C3RB3R ransomware. Federal agencies must remediate the DSM, REDCap and Axis flaws by November 21, 2026, and the LiteLLM flaw by December 2, 2026.

ASUS released Control Center Express v1.7.29 to fix three critical vulnerabilities in its central dashboard for monitoring and remotely managing PC fleets. CVE-2026-114202, rated CVSS 10.0, stems from a hard-coded cryptographic key that lets an unauthenticated attacker forge an authentication token and gain administrator privileges without user interaction. CVE-2026-114204, another hard-coded key issue, lets a low-privileged attacker forge a session token for a recently logged-in account, and CVE-2026-114205 allows a low-privileged user to create an administrator account. ASUS says versions before v1.7.29 are affected, while CVE records list only versions through v1.7.27, leaving v1.7.28 unclear. Users should install v1.7.29, update agents via Select Function → Client Agent Updater and verify versions. ASUS also patched CVE-2026-126574 in Control Center Enterprise and Datacenter, which lets a Viewer-role user change an administrator's password, and CVE-2026-112115, a high-severity local privilege escalation in Armoury Crate affecting versions up to 6.7.2.

Supply Chain Theft and an AI-Driven Intrusion

On November 17, 2026, versions 1.26.1 and 1.26.2 of Microsoft's official agent-framework package on PyPI were published with a credential-stealing payload, as reported by StepSecurity, OX Security and Socket. The open-source toolkit for building AI agents has about 192,534 monthly downloads. The malicious code installs as a .pth startup hook that runs at Python startup without any import, while the legitimate agent-framework-core 1.26.0 dependency keeps the framework working normally. The stealer collects environment variables containing strings such as TOKEN, SECRET, AZURE, AWS and GITHUB, reads the environments of up to 300 other processes, copies .env and credential files for Azure CLI, pip, npm, Git, Docker and Hugging Face, and tries to request fresh Azure managed-identity and CI federation tokens, with 1.26.2 adding GitHub Actions OIDC tokens; there is no confirmation that any were issued. The data is compressed, encoded and sent to a webhook.site endpoint. PyPI quarantined both releases, but 1.26.2 was uploaded after 1.26.1 had already been quarantined, which suggests continued publishing access. Affected users should isolate hosts, rebuild from a clean base, pin to 1.26.0, treat reachable secrets as compromised and review OIDC permissions and managed-identity role assignments. Coder's Eric Paulsen recommends internal mirrors of approved versions, ephemeral build environments, short-lived credentials and egress allow-lists.

The European Commission confirmed an AI-driven cyber-attack that led to services being temporarily taken offline. Spokesperson Thomas Regnier said at a November 19 press briefing that the incident was detected on October 22 and confirmed as an attack on October 26, after which mitigation and an investigation began. Several applications were taken down as a protective measure and most have since been restored. Regnier said he was not aware of any ransomware demand and that no evidence of a data breach affecting Commission staff has been detected, though the investigation is ongoing and no perpetrators have been named. The European Data Protection Supervisor has been notified, and the Commission is working with CERT-EU while preventively checking other systems. Politico reported that affected systems included the Sysper HR platform and a medical reimbursement system, and linked the attack to the Chinese AI penetration-testing tool ARTEX, which CERT-EU and an FBI Flash notice have warned threat actors have used to automate campaigns since at least August.

vCenter Zero-Day, Critical PHP and NGINX Patches Lead the Day · CISO Brief