Post #949250
2026-03-31 15:28 UTC
Nicolas Carlini has been experimenting with what attackers can do with tools like Opus 4.6. He found severe security issues in the Linux kernel that had been lurking since 2003
He describes himself as an LLM skeptic, but concludes that LLMs are already better than the top individual security researchers in the world and soon will exceed the collective capability of the entire security community
https://youtu.be/1sd26pWhfmg
Replies (1)
-
@neilk@xoxo.zone 2026-03-31 17:04
For my own work, which is mostly applications or infra development and often in an open source context, there are *some* things we can do. OS-level guardrails, containerization, capabilities. And… LLMs will help us with that Red Queen’s race. I’m not qualified to discuss what this means for non-upgradeable embedded or legacy systems. Let alone systems with a huge cost of radical change like the Linux kernel. Maybe Tanenbaum will have the last laugh