NT kernel engineer: Linux's messy permission model is a poor fit for AI agents
lauriewired · x · 2026-09-25
Security researcher lauriewired argues the NT kernel, with its coherent day-one security model, beats Linux whose SELinux, Capabilities and Namespaces are bolted-on patches. As users grant AI agents escalating system access, Linux's overlapping mix of UIDs, GIDs, ACLs, cgroups and filesystem modes makes it impossible to answer "what exactly is this agent allowed to do?" NT-style typed resources would enable strong centralized audit trails. He also muses that an "Open NT" in the early 2000s could have let companies like Amazon fork and customize the kernel safely. A greenfield kernel designed in 2026, he says, would look closer to NT or a BSD fork than Linux.
More from AGI Musings
- Primus Society runs a society of autonomous AI researchers, claims 30% training boost — JayAlammar · 2026-09-25
- OpenAI hosts 2026 AGI Economy Conference gathering academia and industry on AI's economic impact — soumitrashukla9 · 2026-09-25
- The Mirror Isn't the Monster: We Keep Redrawing the Line Between Humans and Machines — kevrussell · 2026-09-25
- Why Anthropic Says 'Claude Did' Instead of 'We Did' — A Take on Blame and Hacking Disasters — ryunuck · 2026-09-25
- A 2016 OpenAI Post Resurfaces — Written by Dario, Sam, and Ilya — AashaySachdeva · 2026-09-25
- Rob LeClerc: AI Fear Will Spike Once Humanoid Robots Get Truly Capable — robleclerc · 2026-09-25