Breaking

Anthropic Reports Five Exploit Chains Built by an Unreleased Claude Model

October 28, 2026 at 04:16 EDT

Security
Foundation Models

Anthropic says an unreleased Claude research model produced five working exploit chains after an initial prompt, across Chrome, Safari, Android, Windows and Linux. The demonstrations range from code execution beyond Chrome’s sandbox to remote root access on a Linux server, and the company says vendors have addressed the reported flaws. Anthropic’s report details the findings.

OCT 27, 2026 · ANTHROPIC

Claude links flaws into five exploit chains

An unreleased research model demonstrated attack paths across browsers, phones and operating systems. Vendors have addressed the reported flaws.

working exploit chains

Chrome CVEs assigned

sandbox escapes reported

Linux proof of concept: 53 minutes

The reported timing separates the kernel pointer leak from command execution.

48 min Leak kernel pointer

5 min Run root command

Defensive use

Anthropic recommends scanning code during pull-request review and verifying findings before remediation.

Evidence limit

The report gives no total attempt count or success rate for the five demonstrations.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.