Linus Torvalds Embraces AI in Linux Kernel: "Fork It or Walk Away"

The creator of Linux has drawn a line in the sand — and it's on the side of AI.

In a series of comments during the Linux 7.2 release cycle in August 2026, Linus Torvalds forcefully rebuked critics of AI-assisted code contributions to the Linux kernel, declaring that "Linux is not one of those anti-AI projects" and that AI is "clearly a useful tool." When developers pushed back on AI-generated patches entering the kernel's review pipeline, Torvalds' response was characteristically blunt: "Fork it. Or just walk away."

The exchange marks a watershed moment for the open-source community — and for the broader software industry watching how the world's largest collaborative software project handles the AI revolution.

What Happened

The controversy erupted during the review cycle for Linux kernel 7.2, which Torvalds described as containing "huge" updates — a cadence he attributed partly to the increasing use of AI coding tools by kernel contributors. Several developers raised concerns that AI-generated code could introduce subtle bugs, security vulnerabilities, or maintainability issues that human reviewers might miss.

Torvalds was unmoved. In comments reported by The Register, Tom's Hardware, and Ars Technica, he made clear that the kernel's review process doesn't discriminate based on how code was written — only whether it's correct, well-documented, and meets the project's standards.

"Linux is not one of those anti-AI projects. AI is clearly a useful tool, and we're not going to ban it."

The position is consistent with Torvalds' long-standing philosophy: the kernel accepts good code regardless of its origin. Whether a patch was written by a human at 3 AM or generated by an AI coding agent matters less than whether it works, is readable, and follows kernel conventions.

Why This Matters

1. The Linux Kernel Is the Canary in the Coal Mine

The Linux kernel is the most scrutinized open-source project on Earth. It powers everything from Android phones to cloud servers to supercomputers. If AI-generated code is being accepted into this codebase — under the most rigorous review process in the industry — it sends a signal to every other project: the era of AI-assisted development is not coming. It's here.

2. AI Has Made "Huge" Kernel Updates the New Normal

Torvalds specifically noted that AI tools have contributed to larger, faster release cycles. The kernel's development velocity has increased because contributors can generate, test, and refine code more quickly with AI assistance. Driver updates, bug fixes, and refactoring work that once took weeks can now be accomplished in days.

This isn't speculation — it's observable in the commit logs. The 7.2 release cycle included record-breaking merge volumes, with Torvalds explicitly crediting the improved tooling.

3. The Review Process Is the Safeguard — Not the Authorship

Torvalds' argument rests on a key principle: the kernel's review process is what ensures quality, not restrictions on how code is written. Every patch goes through:

Subsystem maintainer review — domain experts evaluate correctness
Code style checks — automated tools enforce conventions
Testing infrastructure — extensive CI/CD validates changes across architectures
Linus' final pull — Torvalds himself reviews major merges

AI-generated code doesn't bypass any of these steps. It simply increases the volume of code that enters the pipeline.

The Critics' Concerns

Not everyone is convinced. Key arguments from the anti-AI camp include:

Subtle correctness issues: AI models can produce code that looks correct but contains edge-case bugs that only manifest under specific conditions — exactly the kind of bugs that are hardest to catch in review.
Security implications: Kernel code runs at the highest privilege level. A subtle vulnerability introduced by an AI hallucination could have catastrophic consequences.
Maintainability: Code that's AI-generated may be harder for humans to understand and modify later, especially if the original contributor doesn't fully understand what the AI wrote.
Accountability: When a bug is found, who's responsible — the human who submitted the patch, or the AI that generated it?

Torvalds' answer to all of these is essentially the same: the submitter is responsible, and the review process catches problems. If a human submits a bad patch — whether they wrote it by hand or with AI help — they face the same consequences.

The Bigger Picture: AI in Open Source

The Linux kernel isn't the only project grappling with AI contributions. Across the open-source world:

GitHub reported that AI-assisted pull requests increased 340% in 2025-2026
Major projects like Kubernetes, React, and PostgreSQL have all seen growing numbers of AI-assisted contributions
Some projects have explicitly banned AI-generated code (notably, several security-focused projects)
Others have created new labeling requirements, asking contributors to disclose when AI was used

The divergence is creating a two-tier open-source ecosystem: projects that embrace AI as a productivity multiplier, and projects that restrict it in the name of caution.

What This Means for Developers

If you're a developer contributing to open source — or building software in any context — Torvalds' stance offers several takeaways:

AI Is a Tool, Not a Crutch

Torvalds treats AI the same way he treats any tool: it's useful if it helps you write better code, faster. But you remain responsible for what you submit. "I used AI" is not an excuse for a bad patch.

Review Culture Matters More Than Authorship Rules

The kernel's success with AI contributions is possible because of its mature, multi-layered review process. Projects with weaker review cultures may not be able to safely absorb AI-generated code at the same volume.

The Velocity Gap Is Real

Projects that effectively integrate AI tools are moving faster than those that don't. This creates competitive pressure — and may eventually make anti-AI stances unsustainable for projects that need to keep up.

Transparency Is Evolving

While Torvalds doesn't require AI disclosure, the broader industry is trending toward it. Expect norms to settle somewhere between "full disclosure required" and "don't ask, don't tell" over the next year.

The Economic Angle

The AI tools making this possible — from coding agents to code review assistants — aren't free. Developers and organizations are spending real money on API calls to models like GPT-5.6, Claude Opus 5, and DeepSeek V4 to generate and review code.

For cost-conscious teams, the choice of model matters enormously. A kernel contributor running hundreds of AI-assisted coding sessions per week could face very different costs depending on their provider:

GPT-5.6 Sol: $5.00/M input, $30.00/M output (retail)
Claude Opus 5: $5.00/M input, $25.00/M output (retail)
DeepSeek V4 Pro: $0.44/M input, $0.87/M output (retail)

At Qubax, these same models are available at discounted rates — for example, Claude Opus 5 at $4.85/M input and $24.25/M output, saving 3% on every call. For high-volume coding workflows, those savings compound quickly.

Explore all available AI models and pricing → [qubax.ai/models](https://qubax.ai/models)

What's Next

The Linux 7.2 release is expected to be finalized in the coming weeks, and it will likely be remembered as the first major kernel release where AI-assisted development was openly acknowledged as a significant contributor. Torvalds' stance effectively normalizes AI in the kernel — and by extension, in serious software development everywhere.

The critics haven't gone away, and the debate about AI's role in critical infrastructure will continue. But the world's most influential open-source leader has made his position clear: AI is a tool, tools are meant to be used, and if you don't like it, the fork button is right there.

For developers looking to experiment with AI-assisted coding — whether for kernel contributions or everyday workloads — the barrier to entry has never been lower. The question is no longer whether to use AI, but which models and how efficiently.

Start building with AI APIs today → [qubax.ai/docs](https://qubax.ai/docs)

FAQ

Did Linus Torvalds actually say "fork it or walk away"?

Yes. Multiple sources including The Register, Tom's Hardware, and Ars Technica reported Torvalds' comments during the Linux 7.2 release cycle in August 2026, where he told critics of AI in the kernel to "fork it" or "just walk away."

Is AI-generated code now accepted in the Linux kernel?

The Linux kernel has never had a formal policy against AI-generated code. Torvalds' comments clarify that the project evaluates patches based on quality and correctness, not authorship method. AI-assisted contributions are already entering the pipeline.

Does the Linux kernel require contributors to disclose AI usage?

No. Torvalds has not introduced any requirement to disclose whether AI was used in generating a patch. The review process focuses on the code itself, not how it was written.

What are the security risks of AI-generated kernel code?

Critics have raised concerns about subtle bugs, security vulnerabilities, and maintainability issues. The kernel community's position is that its multi-layered review process — including subsystem maintainer review, automated testing, and Torvalds' own review — is sufficient to catch problems.

Which AI models are best for code generation?

Top options in 2026 include GPT-5.6 Sol and Claude Opus 5 for complex reasoning tasks, DeepSeek V4 Pro for cost-efficient coding, and GLM 5.2 for multilingual workloads. You can compare pricing and capabilities at qubax.ai/models.

How much does it cost to use AI for coding assistance?

Costs vary significantly by model. Premium models like GPT-5.6 Sol cost $5/M input and $30/M output at retail, while budget-friendly options like DeepSeek V4 Pro cost under $1/M. Qubax offers discounted rates on all major models — see qubax.ai/models for current pricing.