Anthropic says an unreleased model built working attack chains against Chrome, Safari, Android, Windows and Linux, and 62 Chrome flaws it found are now patched

Anthropic's Frontier Red Team wrote on 27 October that “Over the past few months, we've used an unreleased, internal research model to find critical vulnerabilities” in widely used browsers and operating systems. The post does not name the model. It describes five exploit chains, each of which it says is now fixed:

Chrome: a one-click chain that escapes the browser's sandbox. Anthropic says 62 Chrome CVEs have been assigned and patched as of Chrome 156, all rated critical or high, including 13 sandbox escapes. Safari: a chain that breaks the browser's same-origin rule and exposes history, cookies and logged-in sessions. Android: a chain that takes photos off a locked Pixel 9. Windows: a remote crash in the kernel's QUIC code, CVE-2026-33096. Linux: a zero-click exploit against servers with a multipath TCP listener, which Anthropic says lets an attacker run commands as root without credentials or user interaction.

Anthropic says it gives vendors 90 days to patch, and 45 more days before it discusses details beyond the public advisories. The post does not say how much of each chain the model built on its own, so the model's independence is Anthropic's claim.

The post points security professionals to Anthropic's Cyber Verification Program, which gives vetted defenders access to Claude Opus 5.5, Claude Sonnet 5.5 and Claude Mythos 5.1. It comes a day before Google said its new Gemini 4 Carbon goes first to cyber defenders.

• Confirmed The vulnerabilities were found with an unreleased, internal research model. Anthropic
• Claimed 62 Chrome CVEs have been assigned and patched as of Chrome 156, all rated critical or high. Anthropic
• Claimed The Linux exploit runs commands as root without credentials or user interaction. Anthropic