Google Chrome EUVD-2026-100654

Exploitation requires the attacker to have ALREADY compromised the Chrome renderer process via a separate vulnerability - this is a second-stage sandbox-escape bug, not a standalone remote exploit. Additionally, user interaction is required (UI:R): the victim must load attacker-controlled crafted HTML/content that exercises the ShapeDetection API, triggering the use-after-free. Limiting factors: the mandatory prior renderer compromise, the need for user interaction, and the high attack complexity of successfully grooming the freed object to achieve code execution. The condition is implicitly satisfied on default Chrome configurations below 155.0.8059.79 since ShapeDetection is a standard browser capability, but the chaining requirement substantially raises the bar.

This is a genuinely serious but context-dependent flaw, and the signals mostly agree. The CVSS 3.1 vector (AV:N/AC:H/PR:N/UI:R/S:C/C:H/I:H/A:H = 8.3 High) correctly reflects that this is NOT a standalone remote exploit: the description explicitly requires 'a remote attacker who had compromised the renderer process.' That renderer-compromise precondition is what justifies AC:H, and the sandbox escape justifies the S:C scope change. So despite PR:N, exploitation realistically requires chaining with a separate renderer RCE bug plus user interaction (visiting a crafted page). The lower-urgency signals corroborate this: EPSS is only 0.21% (11th percentile), SSVC reports Exploitation: none and Automatable: no (consistent with no public exploit and the need for a chained precondition), and there is no CISA KEV listing. The one high signal is SSVC Technical Impact: total, which matches the sandbox-escape-to-arbitrary-code outcome. Net: prioritize because it affects a ubiquitous product and is a sandbox escape, but it is not a drop-everything standalone RCE - it is a second-stage component of an exploit chain, and no evidence of active exploitation exists at time of analysis.

An attacker who already controls a compromised Chrome renderer process (via a separate first-stage bug) lures or already has a victim on a page containing crafted content that triggers the use-after-free in the ShapeDetection component. By grooming the freed object, the attacker escalates from the sandboxed renderer to arbitrary code execution in the higher-privileged browser process on the host. No public exploit identified at time of analysis, and the attack requires both the prior renderer compromise and user interaction.