Greg Kroah-Hartman, responsible for supporting the stable and “staging” branches of the Linux kernel, announced that he has stopped accepting patches prepared using AI tools. It is noted that such automatically generated patches contradict the main idea of using the drivers/staging subsystem as a training ground for new kernel developers. The code in drivers/staging is initially positioned as problematic (TAINT_CRAP) and maintainers deliberately do not perform automated cleanup in order to allow novice developers to develop the skill of independently solving minor problems in the kernel.
An exception is made only for AI-generated patches that fix real security problems. At the same time, such driver fixes must be tested on real equipment and provided with a testing report. The author who submitted the patch must be able to prove that the fix is correct. Manual review is necessary because Greg estimates that about a third of the fixes for potential security problems generated by AI models, even the most advanced ones, are incorrect or cause problems.