Torvalds says Linux fixes may be overtaking AI bug reports in one area
The encouraging assessment comes from eBPF’s lead developer. Torvalds says most other kernel areas have not reached the same point.
Loading page…
The encouraging assessment comes from eBPF’s lead developer. Torvalds says most other kernel areas have not reached the same point.
Listen to this story
The report offers a narrowly scoped sign that AI-related bug reports may be becoming more manageable: eBPF co-creator Alexei Starovoitov says fixes in that Linux subsystem are arriving faster than new reports. Linus Torvalds relayed the assessment in the Linux 7.4-rc2 update but cautioned that most kernel areas have not reached the same point. The release candidate’s size also reflected code deferred from the previous candidate, so it does not quantify AI’s contribution or show that AI-driven reports have peaked across Linux.
eBPF runs sandboxed programs inside the Linux kernel.
Torvalds called eBPF’s progress a possible sign of a shift, but said not to count on a broader turnaround yet.
Linux 7.4-rc2 was large partly because Torvalds had held code back from the first candidate, which was already the largest in kernel history.
One part of Linux may be fixing problems faster than AI-related bug reports arrive—a tentative improvement, rather than a kernel-wide turnaround. In his Linux 7.4-rc2 update, Linus Torvalds relayed that assessment for eBPF, while warning that most other kernel areas had not reached the same point, according to The Register.
The observation came as Torvalds delivered the second release candidate, an early version of the next Linux kernel. He described it as a large release for two reasons: code he had held back from the first candidate, and what he called “various AI fixes.” The Register covered the update on November 9.
Torvalds attributed the assessment to Alexei Starovoitov, eBPF’s co-creator and lead developer. Short for Extended Berkeley Packet Filter, eBPF allows sandboxed programs—programs running within restrictions—to execute inside the Linux kernel. The claim concerned that subsystem specifically, not the pace of repairs across the entire project.
This time around Alexei claims that on the bpf side at least the fixes are now coming in faster than the new reports, and that sounds lovely if true
Linus Torvalds, in his Linux 7.4-rc2 update, quoted by The Register
The distinction is between the rate of incoming reports and the rate of fixes. Repairs overtaking reports would be an encouraging balance, but would not, by itself, mean fewer problems were being discovered. Torvalds’s observation was about the relationship between those two streams, rather than a claim that AI had stopped finding flaws.
Nor was the release’s size a clean measure of AI’s contribution. Torvalds said he had omitted some code from the first candidate because he was busy and because that candidate was already the largest in kernel history. Those deferred changes helped make the second candidate large, alongside the AI-related fixes.
For the rest of the kernel, Torvalds’s assessment remained cautious: most areas were not yet keeping ahead of new reports. He called eBPF’s progress a possible sign that the tide was turning, then tempered that hope with a plain instruction: “Don't hold your breath.”
Loading discussion...
Join the conversation
Explain what keeping pace would—or would not—tell you.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.