Broadcom has released fixes for two serious vulnerabilities in VMware Workstation and VMware Fusion, including a critical flaw that can allow an attacker with administrative privileges inside a virtual machine to execute code on the host operating system. The most severe issue, CVE-2026-59346, is an integer-overflow vulnerability in the VMXNET3 virtual network adapter and carries a CVSS score of 9.3. According to Broadcom, a malicious actor with local administrative privileges inside a VM using VMXNET3 may be able to exploit the flaw and execute arbitrary code on the host. The second vulnerability, CVE-2026-59347, is a stack-based buffer overflow in HGFS that can allow an attacker with local administrative privileges inside the guest to execute code as the VMX process running on the host. Both flaws affect VMware Workstation and VMware Fusion versions 25H2 and 26H1 and are fixed in version 26H1u1. Broadcom says there are no workarounds, making patching the only complete remediation currently available. 

The critical point is that these are not merely vulnerabilities inside a guest virtual machine. They affect the isolation boundary between the guest and the host. Virtualization security depends on the assumption that even if an attacker completely compromises a guest operating system, the host remains protected. That assumption is fundamental to malware-analysis environments, developer workstations, security testing labs and any system where untrusted workloads are deliberately isolated inside virtual machines. CVE-2026-59346 directly challenges that model because an attacker who already controls the guest can potentially turn a guest-level compromise into host-level code execution. Once that happens, the virtual machine stops being a containment mechanism and becomes a stepping stone.

CVE-2026-59346 is particularly significant because it affects VMXNET3, VMware’s high-performance virtual network adapter. Virtual network devices are inherently sensitive attack surfaces because they sit directly between code controlled by the guest and privileged virtualization components on the host. The guest is expected to send packets, descriptors and other networking data through that interface continuously. If malformed values can trigger an integer overflow, the attacker may be able to manipulate memory inside the host-side component processing those requests. The vulnerable interface is therefore not an optional administrative service that defenders can simply firewall off. It is part of the normal mechanism used by the VM to communicate with the network.

That architectural position is what makes virtual-device vulnerabilities so interesting to attackers. A compromised guest can control many of the values sent through virtual network adapters, storage controllers, graphics interfaces and shared-filesystem components. These interfaces are designed to process input originating from the guest, which means they must treat that guest as potentially hostile. The hypervisor boundary should therefore be designed with the same assumption that a network service makes about traffic arriving from the internet: all input is untrusted until proven otherwise.

CVE-2026-59347 illustrates the same problem through a different interface. The flaw exists in HGFS, VMware’s host-guest file-sharing mechanism, and involves a stack-based buffer overflow. Broadcom says successful exploitation can allow an administrator inside the guest to execute code as the VM’s VMX process on the host. Although its CVSS score of 8.1 is lower than CVE-2026-59346, the underlying security issue is still serious because it crosses from guest-controlled input into a privileged host process.

Shared-folder functionality deserves particular security attention because convenience features often create direct communication channels between isolation domains. Developers understandably want to move source code and artifacts easily between a host and VM, and analysts want to transfer samples and reports from sandbox environments. But every such feature increases the number of interfaces through which a malicious guest can interact with the host. In high-risk environments, shared folders, clipboard sharing, drag-and-drop and unnecessary virtual devices should therefore be disabled when they are not required.

The requirement for administrative privileges inside the guest should also not be used to dismiss these vulnerabilities. Attackers rarely begin an intrusion with complete control of the hypervisor host. A more realistic chain is initial compromise of the guest through phishing, malicious software or another vulnerability, followed by privilege escalation inside the VM and then exploitation of the virtualization layer. If the guest is deliberately running malware for analysis, the situation is even more obvious: the analyst may intentionally execute attacker-controlled code with administrative privileges inside the virtual machine because the entire purpose of the VM is to contain it.

This means security laboratories and malware-analysis environments should treat CVE-2026-59346 as particularly urgent. A researcher may reasonably assume that executing malware as administrator inside a Windows VM is safe because the guest is disposable and isolated. A reliable VM escape changes that threat model completely. The malware could potentially use the vulnerable virtual device to execute on the researcher’s host, where source code, credentials, browser sessions, VPN access and other sensitive information may exist. Virtualization therefore reduces risk only while the boundary itself remains trustworthy.

Developer environments face a similar issue. VMware Workstation and Fusion are frequently used to create development, testing and compatibility environments on employee laptops. Those VMs may contain software downloaded from external repositories, experimental builds or third-party components. A compromised development VM can therefore provide an attacker with an additional opportunity to reach the underlying workstation, where Git credentials, signing keys, cloud tokens and corporate authentication sessions may be considerably more valuable than anything inside the original guest.

The vulnerabilities also demonstrate why local attack vectors can still receive extremely high severity scores. CVE-2026-59346 has a CVSS score of 9.3 even though Broadcom describes the attacker as requiring local administrative privileges inside the VM. The reason is that the security scope changes. The attacker starts with privileges in one security domain, the guest, and gains code execution in another, the host. In virtualization environments, crossing that boundary is exactly the event the architecture is supposed to prevent.

Organizations should therefore prioritize upgrading VMware Workstation and Fusion to 26H1u1 wherever affected versions are present. Broadcom explicitly states that no workarounds are available for either vulnerability. Workstation 25H2 and 26H1 and Fusion 25H2 and 26H1 are affected, regardless of the host operating system covered by the advisory. Administrators should verify the installed version rather than merely assuming that desktop virtualization software updates automatically, because developer and laboratory systems have a remarkable ability to escape normal patch-management processes while remaining extremely important.

Asset inventory may be more difficult than it first appears. Workstation and Fusion are frequently installed by developers, security researchers or technical teams outside standard server-management workflows. Organizations should therefore search endpoint inventories for VMware executables and installed-package records rather than relying exclusively on centrally managed application lists. Any workstation used for malware analysis, vulnerability research or execution of untrusted workloads should receive priority.

Until updates are deployed, reducing unnecessary guest-to-host integration can help lower attack surface even though Broadcom does not consider any configuration change a complete workaround. Shared folders should be disabled if they are unnecessary, and organizations should review whether every VM genuinely needs the VMXNET3 adapter. However, because Broadcom says there are no supported workarounds for the vulnerabilities, such measures should be considered only supplementary risk reduction rather than alternatives to patching.

Security teams should also think about post-exploitation visibility. A guest-to-host escape may manifest as unusual behaviour in the VMX process, unexpected host-side child processes or filesystem modifications originating from virtualization components. Endpoint detection systems running on the host should therefore monitor VMware processes like any other privileged application rather than implicitly trusting them because they are signed virtualization software. A legitimate `vmware-vmx` process suddenly spawning a command shell or accessing unusual directories should receive immediate investigation.

This is another example of why process identity alone is not a sufficient trust signal. If an attacker exploits a memory-corruption vulnerability inside a legitimate VMware process, all subsequent malicious behaviour can initially appear to originate from a correctly signed Broadcom executable. The relevant security question becomes what the process is doing rather than whether the executable itself is authentic.

The timing of the disclosure also deserves attention because VMware infrastructure has been receiving sustained attacker interest. In July, Broadcom patched multiple critical VMware vulnerabilities, including CVE-2026-59309 and CVE-2026-59310 in vCenter and CVE-2026-47876 in VMXNET3 on ESX. Threat actors began actively exploiting CVE-2026-59310 within days of disclosure, and researchers later estimated hundreds of victim systems across dozens of countries. There is currently no evidence that CVE-2026-59346 or CVE-2026-59347 are being exploited in the wild, but the recent history demonstrates why organizations should not assume VMware vulnerabilities will remain academic for long.

It is also notable that VMXNET3 has appeared in multiple VMware security advisories. Broadcom patched CVE-2026-47876 earlier this year, an out-of-bounds write vulnerability affecting the VMXNET3 virtual network adapter in VMware ESX that could similarly allow an administrator inside a VM to execute code on the host. The recurrence does not imply that the vulnerabilities are identical, but it highlights how difficult virtual-device emulation is to secure. These components process complicated data structures from an adversary-controlled guest while operating close to one of the most sensitive privilege boundaries in the system.

For organizations relying heavily on desktop virtualization, the broader response should therefore include reducing the number of unnecessary virtual devices presented to guests. Every emulated NIC, USB controller, graphics adapter, shared-filesystem feature and integration service expands the attack surface available to a compromised VM. High-risk workloads should receive the minimum virtual hardware required to function.

Isolation between host activities also matters. A workstation used to analyze hostile software should ideally not simultaneously hold privileged corporate sessions, production credentials and sensitive source code. Virtualization is an important isolation technology, but it should not be the only boundary protecting high-value assets. Dedicated analysis hosts, restricted network access and separate administrative environments can reduce the impact if a future guest-to-host escape succeeds.

CVE-2026-59346 and CVE-2026-59347 ultimately demonstrate a fundamental truth about virtualization security.

A virtual machine is not secure merely because it is virtual.

Its security depends on the correctness of every interface through which the guest communicates with the host.

When one of those interfaces contains a memory-corruption vulnerability, an attacker who owns the guest may be able to cross the very boundary the hypervisor was supposed to enforce.

That is why VM escape vulnerabilities deserve disproportionate attention: the guest can be disposable, but the host usually is not.

Broadcom has released security updates for two security flaws impacting VMware Workstation and Fusion, including one critical bug that could result in arbitrary code execution under certain conditions. The vulnerability, tracked as CVE-2026-59346 (CVSS score: 9.3), is an integer-overflow vulnerability that a local attacker with elevated privileges can exploit to run arbitrary code. "A

Source: Critical VMware Workstation and Fusion Flaw Lets VM Admins Execute Host Code via The Hacker News — published 05 Sep 2026.