Linux kernel developers propose removing legacy code due to LLM-generated bug reports

The Linux kernel community is considering removing multiple legacy subsystems from the kernel tree as a response to the increasing volume of security bug reports generated by large language models. According to an LWN.net article posted on April 22, 2026, the proposed removals target code that has become a maintenance burden due to AI-generated reports.
Targeted subsystems for removal
The specific code proposed for removal includes:
- ISA and PCMCIA Ethernet drivers
- A pair of PCI drivers
- The ax25 and amateur radio subsystem (including NET/ROM and ROSE protocols)
- The ATM protocols and drivers
- The ISDN subsystem
The amateur radio protocol implementation and all associated hamradio device drivers are specifically mentioned as having been "a huge bug/syzbot magnet" for a long time. The article states that since nobody stepped up to help deal with the influx of AI-generated bug reports, moving this code out of the kernel tree is seen as necessary to "protect our sanity."
Community discussion reveals context
The comments section reveals important context about this decision. One commenter notes that the title might be slightly ambiguous - it's not that LLM-created security reports helped remove the code, but rather that the increasing number of these reports drove the decision to remove the code.
Another perspective in the discussion suggests the real problem is unmaintained code in large projects like the kernel that has been "pretending to be maintained by being part of some large project instead of being a separate project where the unmaintained status would have been visible years ago."
The discussion also touches on practical considerations:
- Some of this code supports older hardware where changes are mainly due to kernel infrastructure evolution rather than new features
- There's debate about whether newer kernels can even run on hardware that uses PCMCIA Ethernet cards (with one commenter noting that PCI adapters with PCMCIA slots exist)
- Industrial equipment with decades-long lifespans might still use PCMCIA interfaces
- The cost of maintaining legacy code falls on volunteers when manufacturers don't provide ongoing support
This situation highlights how AI tools are changing the software maintenance landscape. When LLMs can generate security reports at scale, unmaintained or legacy code becomes more visible and burdensome to maintainers who must triage these reports.
📖 Read the full source: HN LLM Tools
👀 See Also

Claude Code users hitting usage limits faster than expected, bugs suspected
Anthropic acknowledges Claude Code users are exhausting quotas 'way faster than expected,' with users reporting maxed-out limits within hours. Suspected bugs in prompt caching may be inflating costs by 10-20x, and downgrading to version 2.1.34 reportedly helps.

Formally verified 3D CSG mesh intersection: trust 93 lines of spec, not AI code
First formally verified 3D CSG mesh intersection in Lean 4. Trust 93 lines of spec, not 1000+ lines of AI-written code. Web demo runs in browser.
Claude Code v2.1.237: Prompt Caching Fix for LLM Gateways + Built-in Concise Output Style
Claude Code v2.1.237 fixes prompt caching for LLM gateway/custom base URL users and adds a built-in 'Concise' output style that skips preamble and narration.

DeepSeek V4 Flash Cost Breakdown: Cache Hit Rate and Price Ratio Explained
DeepSeek V4 Flash costs 0.0066x per agentic task compared to Opus 4.7, driven by 97% cache hit rate and 0.02 cache read-write price ratio.