```
The Linux kernel's staging area will now reject patches generated by large language models, according to a report from Phoronix. The policy, announced by lead maintainer Greg Kroah-Hartman, carves out a specific exception for genuine security fixes.
The report states that Kroah-Hartman cited an "onslaught" of LLM-produced patches as the reason for the change. The staging tree is traditionally an area for new contributors to submit experimental or incomplete drivers for mentorship. This automated influx, the report indicates, has disrupted that workflow by flooding maintainers with low-quality code.
Kroah-Hartman, who reportedly uses AI tools himself, is drawing a distinction between raw AI submissions and AI-assisted development. The new rule targets uncurated, machine-generated patches, not the use of language models as a development aid by humans. When a developer uses AI to help write or review code, the final patch is still considered a human submission.
The security exception is notable. Patches for confirmed, valid security vulnerabilities will still be accepted regardless of their origin. This pragmatic approach acknowledges the value of a working fix for a critical flaw. However, it shifts a significant verification burden to maintainers, who must carefully audit these submissions to ensure they are legitimate and not speculative or harmful.
The policy change sets a precedent for how open-source projects might manage the integration of AI-generated code. As such tools become more common, balancing contribution quality with accessibility remains a key challenge. The Linux kernel's approach focuses on accountability and the clear purpose of the specific contribution area.
Looking forward, the community faces questions about scalability. Similar filtering may be needed if LLM submissions begin impacting important Linux kernel subsystems beyond staging. Standardized methods for verifying AI-assisted security patches will also need to be developed to prevent regressions. Ultimately, the update reinforces that while automation can accelerate development, human oversight is essential for maintaining reliable software.
據 Phoronix 報導，Linux 核心的暫存區將開始拒絕大型語言模型產生的補丁。該政策由主要維護者 Greg Kroah-Hartman 宣布，並為真正有效的安全修復設立了具體例外。
報導指出，Kroah-Hartman 以 LLM 產生補丁的「大量湧入」作為政策變更的理由。暫存區傳統上是供新貢獻者提交實驗性或未完成驅動程式以獲取指導的區域。報導指，這類自動化提交淹沒了維護者，導致品質低劣的代碼充斥，干擾了原有工作流程。
據報導，本人亦使用 AI 工具的 Kroah-Hartman 劃清了原始 AI 提交與 AI 輔助開發之間的界線。新規則針對的是未經篩選的機器生成補丁，而非人類使用語言模型作為開發輔助工具。當開發者使用 AI 協助編寫或審查代碼時，最終補丁仍被視為人類提交。
安全漏洞例外尤為值得注意。針對已確認的有效安全漏洞的補丁，無論來源為何，仍會被接受。這種務實做法承認了對關鍵缺陷的有效修正方案的價值。然而，這也將重大的驗證責任轉移到了維護者身上，他們必須仔細審計這些提交，以確保其合法且非推測性或有害。
這項政策變更為開源專案如何管理整合 AI 產生的代碼樹立了先例。隨著這類工具變得更普遍，在貢獻品質與可及性之間取得平衡仍是關鍵挑戰。Linux 核心的方法著重於責任歸屬及特定貢獻區域的明確目的。
展望未來，社群面臨可擴展性的問題。如果 LLM 提交開始影響暫存區以外的 Linux 核心重要子系統，可能需要類似的過濾機制。同時，也需制定驗證 AI 輔助安全補丁的標準化方法，以防止迴歸問題。歸根結底，此更新重申了一點：雖然自動化能加速開發，但人為監督對於維持可靠軟件仍然至關重要。