Torvalds: AI is fine for competent devs, but spare us the 'drive-by' patches

Linus Torvalds argues that AI belongs in software development when used by people who understand the code and can judge the results. The Linux kernel community has dropped its requirement that contributors disclose AI assistance, reflecting the view that the tool’s use matters less than the quality of submitted work. Torvalds rejects blanket objections to AI-generated commit messages and sees AI as a potentially valuable aid, including a way for newcomers to discover the enjoyment of programming. He distinguishes such learning from “drive-by” contributions: patches submitted by people who neither understand the code nor intend to become active, responsible members of the community.

AI is creating two distinct pressures on kernel maintainers. It can uncover genuine bugs and produce useful security reports, but the volume of findings adds work for an already stretched team; AI-generated patches from contributors who cannot assess them create a separate quality problem. Maintainer Greg Kroah-Hartman says AI security reports have improved from obvious noise to credible findings. AI can also speed up reverse-engineering hardware behavior, potentially making it easier to write Linux drivers for devices lacking support. The argument is not that AI has no risks - concerns include unreliable outputs and disregard for the rights attached to training material - but that its harms stem largely from how people use it. Competent developers can make it productive; uncomprehending submissions should be discouraged.