Google Chrome CVE-2026-9918

Exploitation requires the victim to load attacker-controlled HTML in a vulnerable Chrome build (<148.0.7778.216) with WebGPU enabled (the default on supported GPUs/platforms) so that the Tint shader compiler processes attacker-supplied WGSL. The CVSS vector confirms UI:R (the user must visit/render the page) and PR:N (no authentication to the browser process). Limiting factors: user interaction is mandatory, hosts without WebGPU support or with WebGPU disabled via policy/flag are not exposed, and Chrome's silent auto-update narrows the exposure window for consumer installs. Enterprise environments that defer Chrome updates, kiosks, and users on Chromium-derivative browsers that have not yet rebased the fix remain at elevated risk.

Signals are mixed but lean toward 'patch promptly, not emergency.' CVSS 9.6 is driven by network attack vector, low complexity, no privileges, and scope change with full CIA impact - sandbox escape is inherently high-severity. However, EPSS is only 0.03% (11th percentile) and CISA SSVC reports Exploitation: none and Automatable: no, while Technical Impact is total. There is no KEV listing and no public POC identified, so this is not currently being weaponized at scale. User interaction (UI:R) is required - the victim must load attacker-controlled HTML - which limits drive-by mass exploitation but is trivial via phishing or malvertising. The combination (high theoretical impact, low current exploitation signal, browser auto-update available) makes this a standard high-priority browser patch rather than an emergency response item; however, sandbox escapes are historically chained with renderer RCE bugs by sophisticated actors, so latency in patching meaningfully raises risk for targeted users.

A user is lured (phishing email, malvertising, watering-hole) to a page hosting attacker-controlled JavaScript that requests a WebGPU adapter and submits a crafted WGSL shader. When Chrome's Tint component compiles the shader, the inappropriate implementation is triggered, allowing the attacker to cross the sandbox boundary and execute code in a higher-privileged Chrome process - typically chained with a follow-on payload for persistence or lateral movement. No public exploit identified at time of analysis.