Chrome CVE-2026-4676

Victim must open a crafted HTML page in Google Chrome versions prior to 146.0.7680.165. Remote unauthenticated attacker can deliver the malicious page via link or email. No special Chrome configuration required; default installations are vulnerable.

The real-world risk is substantial despite missing CVSS and EPSS metrics. The attack vector is network-based (remote HTML delivery), requires no user interaction beyond visiting a malicious site, and the impact is critical—sandbox escape enables privilege escalation from renderer process to system or user context. The vulnerability affects all Chrome users below version 146.0.7680.165, representing millions of endpoints. The absence of publicly documented proof-of-concept or active exploitation in the wild (KEV status unknown) suggests this is a proactively patched zero-day or near-zero-day. However, given Chromium's high severity classification and the sandbox escape primitive, this warrants immediate patching before threat actors develop reliable exploits. The combination of network attack vector, low complexity, and critical technical impact (code execution + privilege escalation) places this in the highest practical risk tier regardless of CVSS scoring delays.

An attacker hosts a malicious website containing a crafted HTML page with specially constructed WebGPU commands that trigger the use-after-free condition in Dawn's GPU command buffer processing. When a Chrome user visits the site, the malicious JavaScript executes in the renderer process sandbox, sending GPU commands that manipulate freed memory objects to corrupt critical data structures. By chaining this with known sandbox escape techniques, the attacker reads/writes outside the sandbox boundary, achieving local code execution at the user's privilege level—potentially leading to credential theft, malware installation, or further system compromise. No publicly documented proof-of-concept is known at the time of analysis, but the specificity of the vulnerability and Chromium's proactive disclosure suggest this was likely fixed before reliable exploitation code entered the wild.