Anthropic says Claude wrote exploits for 9.8-rated Linux kernel flaw

A frontier model producing working exploit code against critical infrastructure sharply raises AI security stakes.

A Linux kernel TCP flaw tracked as CVE-2026-98432, rated 9.8 CVSS, affects Ubuntu, RHEL and other distributions. Anthropic said its Claude model wrote an exploit using the flaw, and it uses Claude to flag safeguard misses.

Domain

general

Read it at The Hacker News →

Coverage: 3 sources, in order

1. 29 Oct 15:57 The Hacker News Read the article →
2. 29 Oct 18:15 Dark Reading Read the article →
3. 29 Oct 18:36 RuntimeWire Read the article →

← Front page