Lynch detailed prompting Claude with Linux kernel source sections, after which the model identified a subtle bug missed by human auditors for 23 years (https://mtlynch.io/claude-code-found-linux-vulnerability/). The report cites confirmation from kernel maintainers that the issue was present in code introduced in the early 2000s.

Lynch's post references the specific kernel component analyzed and notes the AI's step-by-step reasoning that isolated the flaw, consistent with details in the Hacker News discussion thread (https://news.ycombinator.com/item?id=47633855). A 2023 Stanford study on LLM code analysis previously documented similar AI capabilities in locating edge-case errors in legacy C codebases.

Lynch reported the finding led to a kernel patch submission, with primary sources limited to his experiment documentation and linked kernel commit records.