Kernel staging maintainer will reject LLM-generated patches

Greg Kroah-Hartman will automatically reject large-language-model-generated patches for the Linux kernel's drivers/staging subsystem, with a narrow exception for hardware-tested security fixes.

Staging exists primarily so new contributors can learn kernel development, he wrote on the linux-kernel list. It is full of low-hanging cleanups and API work in code no one should depend on (loading it sets the kernel's TAINT_CRAP flag). Maintainers deliberately leave those issues in place rather than bulk-fix them with tools, because absorbing automated cleanups would remove the on-ramp the tree is meant to provide.

"Anyone attempting to use a LLM to cleanup or 'fix' any code in drivers/staging/ is EXPLICITLY defeating the whole purpose of it existing in the first place," Kroah-Hartman wrote. LLM-written submissions are usually obvious; concealing their origin will be treated as an attempt to trick maintainers.

One exception stands: an LLM-spotted security issue may still be sent, but only after the submitter has tested the fix on the actual hardware for that driver. Even strong current tools, he noted, produce results that are flat wrong or harmful roughly a third of the time.