Google Chrome CVE-2026-17801

Exploitation requires the victim to open an attacker-controlled HTML page in an unpatched Chrome build before 151.0.7922.72 - user interaction (UI:R) is mandatory, so there is no zero-click or purely passive path. The crafted page must exercise the vulnerable ANGLE/WebGL code path, meaning the target must have ANGLE-backed graphics rendering active (the default; disabling WebGL or hardware acceleration removes the attack surface). Limiting factors: the outcome is described as a 'potential' sandbox escape rather than confirmed reliable code execution, no authentication is needed (PR:N) but the user must navigate to the malicious content, and only Chrome versions prior to the fixed 151.0.7922.72 build are affected.

The signals here are in tension and should be weighed carefully. The provided CVSS 3.1 base score of 9.6 (AV:N/AC:L/PR:N/UI:R/S:C) reflects a scope-changing sandbox escape reachable over the network with no privileges and only user interaction (visiting a page). However, Google labeled the Chromium security severity as 'Medium', and the description qualifies the impact as 'potentially perform a sandbox escape' - suggesting the escape is not fully proven or requires chaining. EPSS is low at 0.21% (11th percentile), the CVE is not on CISA KEV, and no public exploit is identified, so this is not evidence of widespread or imminent exploitation. The 9.6 CVSS therefore appears to overstate near-term operational risk relative to the vendor's Medium rating; treat it as a serious memory-safety browser bug worth prompt patching in Chrome's normal update cycle rather than an emergency zero-day. Note the conflict explicitly: high CVSS and scope change versus vendor 'Medium' severity and low EPSS - the word 'potentially' implies the sandbox escape may not be reliably weaponizable from this bug alone.

An attacker hosts or injects a crafted HTML page containing malicious WebGL/graphics content and lures a victim to open it (phishing link or malvertising). When the page renders, ANGLE performs an out-of-bounds read and write while processing the crafted graphics commands, corrupting memory in a way that could allow the attacker to escape the renderer sandbox. No public proof-of-concept is currently identified, and reliable exploitation likely requires additional chaining given the vendor's Medium severity and the 'potentially' qualifier.