Kernel code removals driven by LLM-created security reports
TL;DR Highlight
LLM-generated security reports are pushing the Linux kernel to remove old networking code.
Who Should Read
Backend or platform developers overwhelmed by AI-generated bug and vulnerability reports on open-source or legacy code
Core Mechanics
- Kernel is deleting code over LLM reports
- Several removal efforts are underway, mostly in the networking subsystem.
- Removing code is the alternative to handling the growing flood of LLM-generated security reports.
- Targets are old hardware and protocols
- ISA and PCMCIA Ethernet drivers and a pair of PCI drivers are on the list.
- The ax25/amateur radio subsystem, ATM protocols and drivers, and ISDN are also included.
- Amateur radio was already a bug magnet
- One patch removes AX.25, NET/ROM, ROSE and all hamradio drivers from the tree.
- Its description says this code has long been a huge bug and syzbot (kernel fuzzing bot) magnet.
- No one stepped up to help
- The patch says nobody helped deal with the influx of AI-generated bug reports.
- So the code moves out of tree to protect maintainers' sanity.
Community Reactions
- Pushback that old hardware is still used
- Commenters noted PCI adapters with PCMCIA slots let newer kernels still run PCMCIA cards.
- Some pointed out industrial equipment lasts for decades and uses many PCMCIA card types.
- Opinion that vendors should carry maintenance
- Some argued users of such equipment should pay Linux vendors to maintain support.
- Others said there is almost no demand to run the latest kernels on decades-old hardware.
- Opinion that in-tree placement hid the problem
- One comment said unmaintained code pretended to be maintained by living inside a big project.
- Another said LLMs merely shine a light on pre-existing problems.
- Proposal for maintenance levels on modules
- One idea was a maintenance-level metric so distros choose what to include by default.
- Several suggested moving neglected code to a separate tree or userspace.
How to Apply
- Count reports per legacy module today
- Tally recent security and bug reports per module in your repo.
- List modules with many reports and no owner.
- Label unmaintained modules
- Note owner and support level in README or CODEOWNERS.
- Move ownerless modules out of the default build and make them opt-in.
- Set intake rules for LLM reports
- Require reproduction steps and a proposed patch in every report.
- Auto-label submissions that fall short as on hold.
- Announce removals before deleting
- Put planned removals in release notes first.
- Point users to a separate repo or userspace replacement.
Try It
# Paste into Claude or ChatGPT as-is (a simplified prompt version of the article's situation)
Below is the list of recent security and bug reports for our repo.
{{report list}}
Make a table per module with:
1. Number of reports
2. Number of people who committed in the last 12 months (see {{commit log}})
3. Recommended action (remove, split out, or opt-in) with a one-line reason
Finally, pick the top 3 modules with no owner and many reports. Don't guess missing info; write 'unknown'.Terminology
Related Papers
The Judge Can Debias Itself: Awareness and Self-Debiasing for Reliable LLM-as-a-Judge
Warning an LLM judge about bias or letting it self-critique improved 20 of 21 model–bias combinations.
90% of US working hours are spent on tasks not covered well by AI benchmarks
Mapping 29 benchmarks to job tasks shows 90% of US working hours lack good coverage
Postgres 19 delayed after AI reviewers find more bugs than humans can fix
LLM code review surfaced so many bugs that Postgres 19 slipped to November and 8-10 features were reverted
Show HN: Gemini 4 Argon plays StarCraft
LLMs wrote StarCraft bots in one hour; GPT-6 Astra scored highest.
A Plausible Explanation Is Enough to Mislead Multimodal Judges
Adding plausible explanations to fixed answers cut MLLM judges' wrong-answer rejection by 12.7–13.8 points.
Neural networks are clean
Looking at neurons directly shows most have one main job, challenging superposition.
Linus Torvalds's first LLM-assisted patch series
Linus posted an LLM-assisted 10-patch series giving an order-of-magnitude read() speedup, with a race already found.
Show HN: Parsimony – an open-source benchmark for how much code AI models add
Counting syntax units in SWE-bench patches ranks which AI models add the least code