Google Chrome CVE-2026-18004

Exploitation has two explicit prerequisites: first, the attacker must have already achieved compromise of the Chrome renderer process through a separate, independent vulnerability - this is explicitly stated in the description and is not a minor prerequisite. Second, the victim must visit or interact with a crafted HTML page, reflecting the UI:R requirement in the CVSS vector. The vulnerability is specific to the Speech component, implying the Web Speech API must be accessible and invocable in the browser context. The CVSS PR:N metric reflects that no authentication to a service or system is required, only web-level access. Exploitation is substantially limited by the renderer-compromise prerequisite, Chrome's existing process sandbox architecture that constrains renderer process capabilities, and the limited confidentiality-only impact with no integrity or availability consequence.

The official CVSS 3.1 score of 4.3 with vector AV:N/AC:L/PR:N/UI:R/S:U/C:L/I:N/A:N suggests a moderate, network-accessible information disclosure. However, this vector materially understates the effective attack complexity: the description explicitly requires prior renderer process compromise - a substantial non-trivial prerequisite - before this vulnerability can be triggered. This creates a significant gap between the theoretical CVSS score and practical exploitability. The Chromium security team itself rated severity as 'Low,' aligning with the constrained real-world impact. The EPSS score of 0.16% (6th percentile) corroborates low exploitation probability. No CISA KEV listing and no public POC further reduce urgency. The disconnect between PR:N in the CVSS vector and the renderer-compromise prerequisite should be noted: PR:N reflects unauthenticated web-level initial access, not the full exploitation chain complexity. This vulnerability is not a priority in isolation but warrants patching to close a potential link in a chained exploit.

An attacker who has separately exploited a Chrome renderer vulnerability (such as a V8 engine or Blink parser bug) during a victim's browsing session then serves a crafted HTML page that invokes the Speech API in a manner that bypasses cross-origin policy checks, allowing the attacker to read data from a different origin - such as a banking tab or authenticated session content - within the same browser instance. No public POC has been identified. This attack path is most plausible as the second stage of a chained exploit where a separate renderer escape provides the necessary foothold.