Claude Mythos autonomously found and chained together several vulnerabilities in the Linux kernel, allowing an attacker to escalate from an ordinary user to complete control of the machine.

“Anthropic's own Frontier Red Team blog and multiple tech publications confirm this capability. A post from Anthropic states Mythos "autonomously obtained local privilege escalation exploits on Linux... by exploiting subtle race conditions and KASLR-bypasses." Reporting from Forbes and VentureBeat specifies that the model chained multiple (two to four) low-severity vulnerabilities to achieve full local privilege escalation, or complete machine control from a user account. Context: This capability is part of a broader set of findings where the model, named Claude Mythos Preview, identified thousands of zero-day vulnerabilities across all major operating systems and browsers. Due to these powerful offensive capabilities, Anthropic has not made the model publicly available, instead providing it to a consortium of tech companies (including the Linux Foundation) under "Project Glasswing" to proactively find and fix security flaws.”