← Back to Feed CACHED · 2026-07-29 17:42:03 · CACHE_KEY CVE-2026-47876
CVE-2026-47876 · CWE-787 · Disclosed 2026-07-29

VMware ESXi VMXNET3 Out-of-Bounds Write VM Escape

ASSESSED — NOISGATE V0.5
Vendor
Reassessed
Verdict:
Do you agree?
01 · The Real Story

A tenant punching through the apartment floor and rewiring the entire building's electrical panel

CVE-2026-47876 is an out-of-bounds write in VMware ESXi's VMXNET3 virtual network adapter emulation code. An attacker who holds local administrative privileges inside a guest VM can craft malicious operations against the paravirtualized VMXNET3 device to corrupt memory in the VMX process running on the ESXi host, achieving arbitrary code execution outside the VM boundary. This is a classic guest-to-host VM escape. Affected versions span ESXi 8.0 (all builds before ESXi80U3k-25595708), ESXi 9.0.x (before 9.0.2.0100-25595025), ESXi 9.1.x (before 9.1.0.0200-25557999), and VMware Cloud Foundation 5.x via async patching. VMXNET3 is the default and recommended virtual NIC adapter for all modern VMware VMs — meaning the vulnerable configuration is near-universal. The bug was discovered by Nguyen Hoang Thach of STARLabs SG and demonstrated as a working exploit at Pwn2Own Berlin 2026 in May 2026, earning $200,000. Broadcom published patches via VMSA-2026-0006 on July 29, 2026.

Broadcom rates CVE-2026-47876 at CVSS 9.3 Critical, and that score is *justified*. The CVSS vector AV:L/AC:L/PR:N/UI:N/S:C correctly captures the scope change — the attack crosses from guest to host — and the full C:H/I:H/A:H impact reflects that a compromised hypervisor means total control over every VM on that host. While the 'Local' attack vector and the requirement for VM admin access might tempt a downgrade, VMXNET3's near-universal deployment, the proven Pwn2Own exploit, and the catastrophic blast radius of hypervisor compromise mean the 9.3 score is appropriate. No vendor workaround exists. This one is as serious as the vendor says.

"VMware ESXi VM escape via VMXNET3 OOB write — hypervisor host compromise from any admin VM."
02 · The Attack Path

5 steps from start to impact.

STEP 01

Establish VM admin foothold

The attacker needs local administrative privileges inside a guest VM running on the target ESXi host. This could come from a compromised workload (web shell, RCE in a tenant application), a malicious insider with vSphere VM console access, or credential theft. In multi-tenant cloud environments, this is the tenant's own VM — the attacker may already be the legitimate admin.
Conditions required:
  • Local administrative access inside a guest VM on the target ESXi host
Where this breaks in practice:
  • Attacker must already have code execution inside a VM — not reachable from the network directly
  • In single-tenant environments, this implies prior compromise of a workload
Detection/coverage: EDR agents inside VMs may detect initial compromise activity; vSphere audit logs capture console/SSH logins
STEP 02

Confirm VMXNET3 adapter is present

The attacker enumerates the virtual hardware of the compromised VM to confirm a VMXNET3 adapter is attached. On Linux guests, lspci | grep -i vmxnet3 or ethtool -i eth0 reveals the driver. On Windows, Device Manager or Get-NetAdapter shows the VMware VMXNET3 adapter. Since VMXNET3 has been VMware's default and recommended adapter since vSphere 5.x, virtually all modern VMs use it.
Conditions required:
  • VM must use VMXNET3 virtual network adapter (default for all modern VMware VMs)
Where this breaks in practice:
  • Only VMs using legacy E1000/E1000E adapters are safe — these represent a tiny fraction of deployments and carry performance penalties
STEP 03

Trigger OOB write in VMXNET3 device emulation

The attacker crafts malicious operations targeting the VMXNET3 paravirtualized device's shared memory / DMA interface between guest and host. The out-of-bounds write corrupts memory in the VMX process — the per-VM userspace process on the ESXi host that handles device emulation. The Pwn2Own exploit demonstrated this is reliable enough for a live competition. No public PoC is available yet, but patch-diffing is actively underway by researchers (Zero Day Engineering has published methodology for prior VMXNET3 bugs).
Conditions required:
  • Kernel-level or admin-level access in guest to interact with VMXNET3 device registers/DMA
Where this breaks in practice:
  • No public exploit code as of July 30, 2026 — but Pwn2Own proof exists and patch-diff window is open
  • Memory layout may vary between ESXi builds, potentially requiring exploit tuning
Detection/coverage: ESXi host-level detection of anomalous VMX process crashes or memory corruption is minimal; no commercial tool reliably detects in-progress VM escape
STEP 04

Achieve code execution on ESXi host

Successful exploitation gives the attacker code execution as the VMX process on the ESXi host, which runs with elevated privileges. From here, the attacker can escalate to root on the ESXi host. This crosses the VM isolation boundary entirely — the attacker is now on the hypervisor.
Conditions required:
  • Successful memory corruption leading to controlled code execution in VMX
Where this breaks in practice:
  • ESXi has limited attack surface hardening (no full ASLR in all configurations, limited seccomp)
Detection/coverage: ESXi syslog may capture VMX process anomalies; no mainstream EDR runs on ESXi itself
STEP 05

Pivot to all co-resident VMs and vCenter

With hypervisor-level access, the attacker can: (1) read memory of all VMs on the host to harvest credentials, (2) modify VM disk images to implant backdoors, (3) pivot to vCenter if the host is managed (using the host's vpxuser credentials), and (4) move laterally to other ESXi hosts in the cluster. In multi-tenant environments, this is a cross-tenant compromise. The Pwn2Own demonstration explicitly included the 'cross-tenant code execution' add-on category.
Conditions required:
  • Root or VMX-level access on ESXi host
Where this breaks in practice:
  • vCenter may have separate credentials if not using linked mode, slowing cluster-wide pivot
Detection/coverage: vCenter audit logs may show unexpected host operations; SIEM correlation of host-level syslog with VM activity anomalies
03 · Intelligence Metadata

The supporting signals.

In-the-wild exploitationNo known exploitation. Broadcom states 'no information to suggest exploitation of these issues has occurred in the wild' as of July 29, 2026.
Proof-of-conceptDemonstrated at Pwn2Own Berlin 2026 (May 16, 2026) by Nguyen Hoang Thach (@hi_im_d4rkn3ss) of STARLabs SG. Full VM escape with cross-tenant code execution, earning $200,000. Exploit code is not public, but Zero Day Engineering has published patch-to-PoC methodology for prior VMXNET3 bugs — expect reproduction within weeks.
EPSS scoreNot yet scored — CVE disclosed July 29, 2026; EPSS typically populates within 30 days. Prior ESXi VM escape CVEs (e.g., CVE-2025-22224) reached top-1% EPSS rapidly.
KEV statusNot listed on CISA KEV as of July 30, 2026. No in-the-wild exploitation reported yet.
CVSS vectorCVSS:3.1/AV:L/AC:L/PR:N/UI:N/S:C/C:H/I:H/A:H9.3 Critical. Scope Changed (S:C) reflects guest-to-host boundary crossing. AV:L because the attacker operates from inside a VM. PR:N is scored from the hypervisor's perspective (attacker is unprivileged to the host).
Affected versionsESXi 8.0 (all builds < U3k), ESXi 9.0.x (< 9.0.2.0100), ESXi 9.1.x (< 9.1.0.0200), VMware Cloud Foundation 5.x, VMware Telco Cloud Platform 5.0.x/5.1.x
Fixed versionsESXi80U3k-25595708, ESXi-9.0.2.0100-25595025, ESXi-9.1.0.0200-25557999; VCF 5.x via async patch KB88287; Telco Cloud via KB449886
Exposure dataShodan/Censys show 30,000–41,500 internet-exposed ESXi instances globally. VMXNET3 is the default adapter on >95% of VMware VMs — the vulnerable configuration is near-universal.
Disclosure timelineMay 16, 2026: Pwn2Own Berlin demo → July 29, 2026: VMSA-2026-0006 published with patches → 74-day disclosure-to-patch window
ReporterNguyen Hoang Thach (@hi_im_d4rkn3ss), STARLabs SG — reported via ZDI/Pwn2Own Berlin 2026
04 · The Call

noisgate verdict.

Final Verdict
= UNCHANGED to CRITICAL (9.3/10)

The single most decisive factor is role identity — VMware ESXi is a hypervisor, and this bug is a proven VM escape, meaning successful exploitation converts a single compromised VM into full host compromise with access to every co-resident workload, their memory, and their disks. The Pwn2Own-demonstrated exploit, near-universal VMXNET3 adoption, and absence of any vendor workaround leave no room to temper the severity below CRITICAL.

HIGH Vulnerability existence and exploitability (Pwn2Own-proven)
HIGH Blast radius assessment (hypervisor compromise = fleet-scale)
MEDIUM Timeline to public PoC (patch-diffing in progress, estimated 2-6 weeks)

Why this verdict

  • Proven exploitability: A working exploit was demonstrated live at Pwn2Own Berlin 2026 — this is not a theoretical bug. The researcher achieved full cross-tenant code execution, confirming the complete attack chain from VM admin to hypervisor control.
  • Near-universal attack surface: VMXNET3 is VMware's default and recommended virtual NIC since vSphere 5.x. Switching to E1000/E1000E is a performance-degrading non-starter for production. >95% of VMware VMs run VMXNET3, so effectively every ESXi host is vulnerable.
  • No vendor workaround: Broadcom explicitly states no workaround exists. The only fix is the patch. Interim mitigations (restricting VM admin, network segmentation) reduce likelihood but do not eliminate the vulnerability.
  • Role multiplier: ESXi is canonically a hypervisor — the highest-value infrastructure role. VM escape on ESXi yields: (a) host compromise → access to all VMs on the host, (b) credential harvesting via VM memory inspection, (c) lateral movement to vCenter and other cluster hosts via vpxuser, (d) cross-tenant breach in multi-tenant/VDI environments. Blast radius is fleet-scale. Per the high-value-role rule, the affected component IS the hypervisor itself, meaning ≥10% of installs (100%, in fact) occupy the high-value role — verdict floor is CRITICAL.
  • Patch-diff window is open: With patches released July 29, 2026 and established patch-to-PoC methodology published for prior VMXNET3 bugs, weaponized exploits could appear within 2-6 weeks. The clock is ticking.

Why not higher?

The score is already at 9.3 Critical — the highest practical rating for a local-vector vulnerability with scope change. A 10.0 would require network-reachable exploitation without any prerequisites, which this bug does not have (it requires VM admin access and local code execution inside a guest).

Why not lower?

Downgrading below CRITICAL would require evidence that the VM-admin prerequisite meaningfully limits exploitation *in the context of a hypervisor*. It does not: in cloud/hosting environments, every tenant IS a VM admin. In enterprise environments, compromised workloads routinely achieve VM-level admin. The blast radius of hypervisor compromise (every VM on the host, potential cluster-wide pivot) is too catastrophic to absorb any friction discount. The hard-rule floor for a hypervisor VM escape is CRITICAL, and no friction factor present here breaks that floor.

05 · Compensating Control

What to do — in priority order.

  1. Restrict VM administrative access immediately — Audit and minimize who holds VM admin/root/SYSTEM credentials across your fleet. Remove standing admin access from VMs that don't require it. Use just-in-time privileged access for VM administration. This reduces the pool of VMs from which the exploit can be launched. Deploy within 3 days per noisgate mitigation SLA for CRITICAL.
  2. Enable vSphere VM encryption and vTPM where feasible — Encrypted VMs add a layer of memory protection that complicates (but does not prevent) host-level credential harvesting after a VM escape. This limits post-exploitation value. Deploy within 3 days.
  3. Segment ESXi management networks aggressively — Ensure ESXi management interfaces (port 443, SSH, DCUI) are on isolated VLANs unreachable from VM guest networks. This limits an attacker's ability to pivot from a compromised host to vCenter or other ESXi hosts. Deploy within 3 days.
  4. Monitor VMX process anomalies on ESXi hosts — Forward ESXi syslog (/var/log/vmkernel.log, /var/log/hostd.log) to your SIEM. Alert on VMX process crashes, unexpected restarts, or core dumps — these may indicate exploitation attempts. Deploy within 3 days.
  5. Prioritize patching hosts running multi-tenant or high-value workloads — If you cannot patch all hosts simultaneously, triage by blast radius: VDI clusters, shared hosting, and hosts running domain controllers, PKI, backup infrastructure, or production databases go first. Begin patch rollout within 3 days, complete within 90 days per noisgate remediation SLA.
What doesn't work
  • Switching from VMXNET3 to E1000/E1000E adapters — while this removes the vulnerable component, it causes severe network performance degradation (loss of TSO, LRO, RSS), breaks VMware best practices, and may destabilize production workloads. Broadcom explicitly recommends patching over adapter swaps.
  • Network-level firewalling or IDS/IPS — this is a local VM-to-host exploit via paravirtualized device memory, not a network attack. No network security control can see or block the malicious device operations.
  • Guest-level EDR/AV — endpoint detection inside the VM cannot observe or prevent the VMXNET3 device-level memory corruption that occurs in the host's VMX process. EDR may detect the initial VM compromise but not the escape itself.
06 · Verification

Crowdsourced verification payload.

Run this script on each ESXi host via SSH (or remotely via esxcli through vCenter). Requires root access. Example: ssh root@esxi-host 'sh /tmp/check_cve_2026_47876.sh'

noisgate-verify.sh
BASHREAD-ONLYSAFE
#!/bin/sh
# CVE-2026-47876 — VMware ESXi VMXNET3 OOB Write (VM Escape)
# Checks ESXi build number against patched builds from VMSA-2026-0006.
# Exit codes: 0=PATCHED, 1=VULNERABLE, 2=UNKNOWN

PATCHED_80=25595708   # ESXi80U3k
PATCHED_90=25595025   # ESXi 9.0.2.0100
PATCHED_91=25557999   # ESXi 9.1.0.0200

# Retrieve version info
VER_LINE=$(esxcli system version get 2>/dev/null)
if [ -z "$VER_LINE" ]; then
  echo "UNKNOWN — unable to query ESXi version. Is this an ESXi host?"
  exit 2
fi

VERSION=$(echo "$VER_LINE" | grep 'Version:' | awk '{print $2}')
BUILD=$(echo "$VER_LINE" | grep 'Build:' | awk '{print $NF}')
PRODUCT=$(echo "$VER_LINE" | grep 'Product:' | awk '{print $2}')

if [ -z "$BUILD" ]; then
  echo "UNKNOWN — could not parse build number."
  exit 2
fi

echo "Host: $(hostname) | Product: $PRODUCT | Version: $VERSION | Build: $BUILD"

# Check for VMXNET3 adapters on any VM
VMXNET3_VMS=$(grep -rl 'vmxnet3' /vmfs/volumes/*//*.vmx 2>/dev/null | wc -l)
echo "VMs with VMXNET3 adapters: $VMXNET3_VMS"

MAJOR=$(echo "$VERSION" | cut -d. -f1)
MINOR=$(echo "$VERSION" | cut -d. -f2)

case "$MAJOR" in
  8)
    if [ "$BUILD" -ge "$PATCHED_80" ] 2>/dev/null; then
      echo "PATCHED — build $BUILD >= $PATCHED_80 (ESXi80U3k)"
      exit 0
    else
      echo "VULNERABLE — build $BUILD < $PATCHED_80 (need ESXi80U3k)"
      exit 1
    fi
    ;;
  9)
    if [ "$MINOR" = "1" ]; then
      THRESHOLD=$PATCHED_91
      LABEL="ESXi 9.1.0.0200"
    else
      THRESHOLD=$PATCHED_90
      LABEL="ESXi 9.0.2.0100"
    fi
    if [ "$BUILD" -ge "$THRESHOLD" ] 2>/dev/null; then
      echo "PATCHED — build $BUILD >= $THRESHOLD ($LABEL)"
      exit 0
    else
      echo "VULNERABLE — build $BUILD < $THRESHOLD (need $LABEL)"
      exit 1
    fi
    ;;
  *)
    echo "UNKNOWN — ESXi major version $MAJOR is outside known affected range (8.x, 9.x)"
    exit 2
    ;;
esac
07 · Bottom Line

If you remember one thing.

TL;DR
This is a day-one hypervisor VM escape with a proven Pwn2Own exploit and no workaround — treat it as a fire drill. Under the noisgate mitigation SLA for CRITICAL, deploy compensating controls (VM admin access restrictions, management network segmentation, ESXi syslog forwarding) within 3 days — by August 2, 2026. Begin patch rollout immediately, prioritizing multi-tenant clusters, VDI hosts, and any ESXi hosts running domain controllers, backup servers, or production databases. Complete patching under the noisgate remediation SLA of 90 days (by October 28, 2026), but realistically push for full coverage within 14-21 days given the Pwn2Own pedigree and active patch-diffing. If you run VMware Cloud Foundation 5.x, apply the async patch via KB88287. There is no workaround — patching is the only remediation. Monitor Broadcom's VMSA-2026-0006 advisory and ZDI disclosures for any escalation to in-the-wild exploitation, which would trigger an immediate patch-now override.

Sources

  1. Broadcom VMSA-2026-0006 Advisory
  2. SecurityWeek — Critical VM Escape Vulnerability Patched in VMware ESXi
  3. The Hacker News — Three Critical VMware Flaws
  4. VMware Security Blog — Pwn2Own Berlin 2026
  5. VMware VCF Security Advisory on GitHub
  6. Security Affairs — Broadcom Patches Critical VMware ESXi Vulnerability
  7. Zero Day Engineering — VMware ESXi VMXNET3 Patch-to-PoC Methodology
  8. Vali Cyber — Pwn2Own Berlin 2026 ESXi VM Escape
Peer Review

What defenders are saying.

Submit a review attribution: handle + country only
0 flags selected · stored anonymously
Validation Results

Crowdsourced verification outputs.

Results submitted by users who ran the verification payload against their environment.