🛡 AI audit of the Linux kernel: a dozen real bug fixes out of 79 “vulnerabilities”
Greg Kroah-Hartman, Linux Foundation Fellow and stable kernel maintainer, on September 22 at Kernel Recipes 2026 in the talk “Security in the LLM age” (the recording was released on September 29) analyzed a Mythos run on the kernel: of 79 claimed vulnerabilities, 24 had no details, 14 were not bugs, 3 were fabricated, and 15 were already fixed (11 by other developers, 4 by Anthropic). About 20 required patches, resulting in “10 real bug fixes.”
🌍 A stream of auto-generated reports shifts triage to maintainers and undermines trust in vendor claims about AI audits, although AI companies are sending patches upstream (Anthropic — four fixes).
👤 Reading about “holes found by AI in the kernel,” or checking code with LLMs yourself — about 20 of the 79 reports turned out to be real; the talk recording is available on YouTube.
Source 1: https://www.youtube.com/watch?v=NnV_cWeoo5Q Source 2: https://news.ycombinator.com/item?id=49929391