Linux Kernel Embraces AI Coding—But Who Takes the Blame?

One‑Sentence Summary

Linux now allows AI‑generated patches, but only under three strict rules that keep humans in charge of signing, attribution, and liability.

Three Ironclad Rules

Rule 1 – AI Cannot Sign "Signed‑off‑by"

❌ AI‑generated Signed‑off‑by → prohibited
✅ Only a human may sign → required

The Signed-off-by tag is the kernel’s legal signature tied to the Developer Certificate of Origin (DCO). Even if the code is 100 % AI‑generated, legal responsibility stays with the human who submits it.

Rule 2 – Must Mark AI Assistance

When an AI tool is used, the commit message must include an Assisted-by tag following this format:

# Example format
Assisted-by: AGENT_NAME:MODEL_VERSION [TOOL1] [TOOL2]

# Real examples
Assisted-by: Claude:claude-3-opus coccinelle sparse
Assisted-by: GitHub-Copilot:gpt-4-turbo
Assisted-by: Cursor:claude-3.5-sonnet sparse

Basic tools such as git, gcc, or editors do not need the tag.

Rule 3 – Human Bears All Consequences

The submitter must fully review and test any AI‑generated code. If problems appear, the human is accountable for:

Bugs

Security vulnerabilities

License non‑compliance

Potential community bans for low‑quality contributions

⚠️ 2021 incident: Researchers deliberately submitted a defective patch to the kernel for “research” and were permanently banned. Maintainers warned that “AI‑written” is not an excuse.

Historical Evolution of AI‑Generated Code

Phase 1 – AI Code as "Garbage" (pre‑2024)

// Typical AI‑generated garbage traits:
// 1. Looks professional but logic is wrong
// 2. Excessive, meaningless comments
// 3. Heavy copy‑paste, unchanged variable names
// 4. Calls to non‑existent APIs

Resulted in the cURL project pausing its bug‑bounty program due to a flood of bogus AI‑generated vulnerability reports.

Phase 2 – Quality Leap (early 2026)

Kernel maintainer Greg Kroah‑Hartman observed a surge of high‑quality AI‑generated security reports, stating that AI tools had made a "qualitative leap."

Community metrics show the shift:

2024 execution score < 50 % – full resistance

Mid‑2025 score ≈ 60 % – discussion begins

Early 2026 score > 72 % – formal acceptance

Current material‑constraint score > 65 % – rules in place

Phase 3 – Formal Policy (12 April 2026)

After intense debate, Linus Torvalds and the kernel team published the AI coding assistant guidelines.

Linus’s Stance: "AI Is Just a Tool"

He does not want the policy to become an "AI manifesto". He treats AI like any other compiler or editor. Future kernels will see many AI‑assisted edge‑case fixes, but core review still relies on human judgment and experience.

Community Debate: Support vs. Opposition

Supporters – Efficiency Advocates

Argument: AI can help close the contributor shortage.

Data: GitHub reports that Copilot boosts basic coding efficiency by >30 %.

Benefits: Lower entry barrier, faster repetitive tasks (API swaps, driver adaptation), and a larger pool of potential contributors (5000 + AI‑tool users worldwide).

Opponents – Quality & Ethics Concerns

Argument: AI‑generated code may dilute expertise and create governance risks.

Data: Q1 2026 saw low‑quality PRs rise from 3 % to 17 % (year‑over‑year).

Data: "warn" integrity ratings increased by 210 % YoY.

Data: AI‑generated code passes audit far less often than human code.

Risk: Over‑reliance could erode foundational skills of new developers.

Practical Impact for Ordinary Users

1. Ubuntu Users

Ubuntu 26.04 LTS (released 23 April) will ship the latest kernel that follows the new AI policy, promising faster bug fixes, more contributor participation, and maintained system stability.

2. Developers’ Career Guidance

┌─────────────────────────────────────────────┐
│            AI‑Era Developer Rules            │
├─────────────────────────────────────────────┤
│ ✓ Learn to use AI tools for speed            │
│ ✓ Keep deep understanding and review ability │
│ ✓ Use AI as an accelerator, not a replacement│
│ ✓ Your name must appear on Signed‑off‑by      │
└─────────────────────────────────────────────┘

3. Open‑Source Ecosystem Direction

Other projects will look to the kernel’s AI policy as a template.

Enterprise software will shape AI adoption strategies based on these rules.

Future legal frameworks may reference the kernel’s liability model.

Key Data Quick‑Reference

AI code execution score: ~72 % (Winzheng Q1 monitoring)

AI material‑constraint score: <65 % (community stats)

Copilot efficiency gain: >30 % (GitHub official data)

Low‑quality PR share: 3 % → 17 % (2026 Q1 vs 2025 Q1)

Integrity "warn" submissions: +210 % YoY

Ubuntu 26.04 LTS release countdown: 8 days (23 April)

Practical Advice for Submitting to the Kernel

Generate an initial draft with AI, then review every line manually.

Run the full test suite, including static and dynamic analysis.

Clearly add the Assisted-by tag to the commit.

Start with small patches; avoid large core changes initially.

Be prepared to defend your design decisions on the mailing list.

This policy marks a turning point: development moves from pure manual effort to a human‑AI collaboration model, responsibility becomes explicit, and quality control shifts from experience‑driven to rule‑driven, reflecting a mature open‑source community’s response to the AI era.