Linus Torvalds sees signs the AI-driven patch flood may have peaked

For most of this year, Linux kernel maintainers have been drinking from a firehose pointed at them by AI bug hunters. Linus Torvalds now thinks the water pressure may finally be dropping.

With the release of Linux 7.4-rc2 on November 8, 2026, Torvalds flagged an encouraging trend. In the eBPF subsystem, fixes are now landing faster than fresh AI-generated bug reports arrive.

A record release, then a slightly late encore

The first draft of this cycle, 7.4-rc1, was enormous. It contained approximately 19,671 commits, a record for the kernel. The typical rc1 lands somewhere around 15,000.

A meaningful chunk of that growth came from AI. The 7.4 cycle saw 3,008 AI-assisted non-merge commits, which is 2.4 times the figure from the prior cycle. These are code changes where AI tools helped find a bug or propose a fix, with humans still required to review and sign off.

Torvalds described rc2 as a big release, and he gave two reasons. One was a deliberate holdback. He left some code out of rc1 partly because he was busy and partly because rc1 was already the largest such release the kernel had ever seen.

Even so, Torvalds called the rc2 diffstat, the summary of which files changed and by how much, “fairly benign.” The changes were spread across subsystems rather than piling up in one corner.

Where the AI surge came from

One cited inflection point is Anthropic’s Mythos model, launched in April 2026. It improved the accuracy and effectiveness of AI bug detection. Better tools meant more real bugs surfaced, which meant more reports landing in maintainers’ inboxes.

Kernel maintainers had previously complained about being buried in low-quality AI-generated patches. Some compared the experience to a denial-of-service attack, the kind of cyberattack where a server is overwhelmed by so many requests that it cannot do its actual job.

The eBPF data point is notable because of who confirmed it. Alexei Starovoitov, the lead developer of eBPF, backed up the observation that fixes are now arriving faster than new reports. eBPF is a technology that lets small, sandboxed programs run safely inside the kernel. It is widely used for networking, observability and security tooling.

Torvalds’ playbook: let AI in early, keep humans in charge

Torvalds has argued that AI-driven changes should be integrated early in the development phase, not saved up and dumped in later during the stabilization stretch. Front-loading the churn into rc1 explains why that release ballooned. It also helps explain why rc2, despite its size, looked orderly.

Torvalds also continues to frame AI as a productivity tool rather than a replacement for the people who make architectural decisions. Strict human oversight of AI-assisted submissions remains a requirement.