{"slug": "linus-torvalds-uses-ai-for-linux-kernel-code-warns-against-low-quality-patches", "title": "Linus Torvalds uses AI for Linux kernel code, warns against low-quality patches", "summary": "Linus Torvalds posted an experimental series of eight Linux kernel patches on October 10, 2026 to a Git branch named 'noref-filemap-read-with-llm', developed with help from Anthropic's Claude Code, which he said generated code, review notes and first drafts of commit messages while he ran the performance measurements himself. In Torvalds' laptop test, throughput rose from approximately 10 million reads per second to 30 to 33 million and kernel cycles per read fell from approximately 2,950 to approximately 725 by letting threads read cached files without touching shared reference counters; the patches are not merged and could come up in the upcoming 7.4 merge window. At the Maintainers Summit on October 8, 2026, kernel maintainers agreed to drop the mandatory 'Assisted-by: LLM' commit tag and judge code on quality, while Torvalds warned against low-quality 'drive-by' AI-generated patches from people who cannot explain the code they submit.", "body_md": "# Linus Torvalds uses AI for Linux kernel code, warns against low-quality patches\n\nThe Linux creator used Anthropic's Claude Code on experimental kernel patches while drawing a hard line against careless AI-generated submissions\n\nLinus Torvalds, the man who built the Linux kernel and has spent decades rejecting sloppy code in public, is now using an AI coding assistant to help write some of his own.\n\nHe isn’t handing over the keys, though. Torvalds says AI is fine in the hands of competent developers. He has far less patience for what he calls “drive-by” patches from people who don’t understand the code they submit.\n\n## What Torvalds actually built with Claude\n\nOn October 10, 2026, Torvalds posted an experimental series of eight patches to a Git branch named ‘noref-filemap-read-with-llm’. The work was done with help from [Anthropic](https://cryptobriefing.com/markets/anthropic/)’s Claude Code, a large language model tool built for programming.\n\nThe patches target a long-running problem in the kernel’s memory management: page-cache contention. When many threads read the same cached file, they all update shared reference counters, and that bookkeeping slows everything down. Torvalds’ patches let those threads read the cached file without touching the shared counters.\n\nThe numbers from his own laptop test are hard to ignore. Throughput climbed from approximately 10 million reads per second to 30 to 33 million. Kernel cycles per read fell from approximately 2,950 to approximately 725.\n\nTorvalds has said openly that the AI tool helped generate code, review notes and the first drafts of commit messages. He says he handled the performance measurements himself, on his laptop.\n\n### AI, tech, and the markets they move—in one daily briefing.\n\nDaily. Free. Join 34,000+ readers across crypto, finance, and policy.\n\n## The patches are not in the kernel yet\n\nThe work is still experimental. As of the latest reports, it has not been merged into the mainline kernel. The patches could come up for consideration in the upcoming 7.4 merge window.\n\nThe page-cache work isn’t his only AI-assisted project. Torvalds also used an LLM to help refactor the obscure ‘assoc_array’ code. He reported finding more bugs during that effort, some dating back to 2013.\n\n## Kernel maintainers drop the AI label\n\nAt the Maintainers Summit on October 8, 2026, core members of the kernel community agreed to drop the requirement for a mandatory ‘Assisted-by: LLM’ tag in commit messages. Maintainers will judge code on its quality, not on how it was produced.\n\nAround October 9, Torvalds spoke at the Open Source Summit Europe. He described AI as a productivity tool for developers, especially for tasks outside their own areas of strength.\n\n## Two very different AI problems\n\nTorvalds has pointed to two separate AI issues facing the kernel. The first is positive: AI tools can be useful for finding bugs, as his own assoc_array work suggests. The second is the opposite. Low-quality AI-generated patches can pile up in maintainers’ inboxes, sent by people who ran a prompt, got some code back and submitted it without understanding it.\n\nThe distinction is about who is holding the tool. An experienced kernel developer using Claude to speed up work they already understand is one thing. A stranger submitting machine-written fixes they can’t explain or defend is another.\n\n**Disclosure:** This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our\n\n[Editorial Policy](https://cryptobriefing.com/editorial-policy/).", "url": "https://wpnews.pro/news/linus-torvalds-uses-ai-for-linux-kernel-code-warns-against-low-quality-patches", "canonical_source": "https://cryptobriefing.com/linus-torvalds-ai-linux-kernel-code/", "published_at": "2026-10-23 12:04:23+00:00", "updated_at": "2026-10-23 12:11:52.449524+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-tools", "large-language-models", "developer-tools"], "entities": ["Linus Torvalds", "Linux kernel", "Anthropic", "Claude Code", "Maintainers Summit", "Open Source Summit Europe"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/linus-torvalds-uses-ai-for-linux-kernel-code-warns-against-low-quality-patches", "markdown": "https://wpnews.pro/news/linus-torvalds-uses-ai-for-linux-kernel-code-warns-against-low-quality-patches.md", "text": "https://wpnews.pro/news/linus-torvalds-uses-ai-for-linux-kernel-code-warns-against-low-quality-patches.txt", "jsonld": "https://wpnews.pro/news/linus-torvalds-uses-ai-for-linux-kernel-code-warns-against-low-quality-patches.jsonld"}}