Google Chrome — Critical WebML and PrivateAI vulnerabilities expose memory data and enable sandbox escape
Google Chrome — Critical WebML and PrivateAI vulnerabilities expose memory data and enable sandbox escape
AI relevance: These vulnerabilities in Chrome's Web Machine Learning (WebML) and PrivateAI components demonstrate how browser-based AI infrastructure creates novel attack surfaces, exposing memory data and potentially enabling sandbox escape attacks that compromise local AI processing security.
Google Chrome users face immediate security risks from two critical vulnerabilities affecting the browser's AI infrastructure components. CVE-2026-5885 (CVSS 8.8) exposes memory data through WebML operations, while CVE-2026-5874 enables sandbox escape attacks through PrivateAI use-after-free conditions, highlighting the security challenges of browser-based machine learning implementations.
Key vulnerabilities
- CVE-2026-5885: Critical WebML memory exposure vulnerability affecting Chrome, Edge, and all Chromium-based browsers on Windows
- CVE-2026-5874: High-severity use-after-free in PrivateAI component enabling sandbox escape attacks
- Memory exposure: WebML flaw allows unauthorized access to memory buffers during ML inference operations
- Sandbox bypass: PrivateAI vulnerability could enable full system compromise through Chrome sandbox escape
- Widespread impact: Affects all Chromium-based browsers including Chrome, Edge, Opera, Brave, and Vivaldi
- AI infrastructure risk: Both vulnerabilities target Chrome's on-device AI processing capabilities
Why it matters
Browser-based AI processing represents a fundamental shift in how machine learning capabilities are deployed, moving from centralized cloud services to on-device execution. While this improves privacy by keeping sensitive data local, it also creates new attack surfaces within browser sandboxes. These vulnerabilities demonstrate that AI components integrated into browsers require specialized security considerations beyond traditional web security models.
The WebML vulnerability specifically threatens the confidentiality of AI processing by exposing intermediate calculation results and potentially sensitive user data. The PrivateAI sandbox escape risk undermines Chrome's fundamental security architecture, which relies on process isolation to contain malicious code. Successful exploitation could allow attackers to compromise the entire system starting from a seemingly benign web page.
What to do
- Immediate patching: Update Chrome to version 147.0.7727.55 or later and apply Windows security updates
- Enterprise monitoring: Implement enhanced monitoring for WebML API calls and memory access patterns
- AI component auditing: Review browser-based AI features for appropriate security boundaries
- Sandbox hardening: Evaluate additional sandbox protection mechanisms for AI processing components
- Supply chain awareness: Recognize that Chromium vulnerabilities affect multiple browser ecosystems simultaneously
- User education: Train users to recognize suspicious AI feature requests from websites