API
< Back to all clusters
[TECHNOLOGY] · United States · 9 sources

started · updated

Anthropic research model develops functional exploits for major operating systems

An unreleased Claude research model developed functional exploit chains for Chrome, Safari, Android, Windows, and Linux, according to a report by Anthropic’s Frontier Red Team. The research, published on October 27, 2026, demonstrated how AI can autonomously identify vulnerabilities and chain them to execute complex cyberattacks in a laboratory setting.

Key findings include a Chrome exploit chain that bypasses the browser sandbox to execute code, and a Safari chain capable of bypassing the same-origin policy to access sensitive user data like cookies and browsing history. In Android testing, the model successfully extracted photos and personal data from a locked Google Pixel 9 via USB. For Linux, the model exploited a critical kernel flaw (CVE-2026-98432) to obtain root shell access in approximately 53 minutes.

All identified vulnerabilities were reported to the respective vendors and have been patched. Google addressed the Chrome vulnerabilities in version 156, and Apple released fixes for the Safari and iOS/iPadOS flaws. The research aims to use AI to strengthen software defenses through formal verification.

Entities

Anthropic · Apple · Chrome · Claude · Google · Linux · Microsoft · windows

Claims

What the coverage asserts, and how many sources carry each claim.