Hardware Ready Ready to Bench?
The Linux Foundation has highlighted a problem that has been affecting the kernel mailing lists for months: the increasing number of patches and security reports generated with artificial intelligence is overloading open source project maintainers.
To address this situation, an initiative was launched in March 2026, entrusted to Alpha-Omega and the Open Source Security Foundation, funded with $12.5 million by Anthropic, AWS, GitHub, Google, Microsoft, and OpenAI. The goal is to provide triage tools to maintainers, who struggle to verify the flow of automatic results with current resources.
Linux Foundation: why it matters
According to maintainer Willy Tarreau, the kernel’s private security list went from two or three weekly reports a couple of years ago to five to ten per day in early 2026. In May, with the announcement of the Linux 7.1 release candidate, Linus Torvalds described that list as “almost entirely unmanageable” due to duplicate reports generated by models.
Since then, kernel documentation has required researchers to make public discoveries obtained with AI assistance. However, Torvalds has never opted for a total closure. In July, responding to restrictive proposals, he stated that “Linux is not one of those anti-AI projects” and that anyone who disagrees can create a fork or abandon the project.
For the kernel creator, AI is “a tool like any other,” useful for identifying bugs even if it increases the review load. The stance hardened on October 15, when Torvalds called bans on AI-assisted code nonsensical, comparing them to banning compilers or text editors.
The kernel has therefore withdrawn the Assisted-by tag, which contributors were supposed to add to commits: in version 7.3, it appeared in only 10% of patches made with model assistance. At the October 8 Maintainers Summit, it was reiterated that low-quality contributions will be placed at the bottom of the queue and the project does not intend to further simplify entry for newcomers.
What changes and what are the effects
Not all maintainers follow the same line: the staging area has started rejecting LLM-generated patches except for real security fixes, while Johannes Berg, head of wireless, announced that he will ignore incorrect automatic proposals after a three-second review. The fundamental principle remains that code responsibility lies with whoever signs it, not the model that wrote it.
In the coming months, the focus will therefore be on filtering tools rather than bans.
Source and further reading on Linux Foundation: original article.
* Content created with the assistance of artificial intelligence systems.