Linus Torvalds, the visionary creator of Linux, recently offered a candid glimpse into his evolving relationship with artificial intelligence at the Open Source Summit North America. He openly admitted to a “love-hate relationship” with AI, acknowledging its revolutionary potential while highlighting the significant social and security challenges it introduces, particularly within the open-source ecosystem. Torvalds firmly believes AI is a powerful tool reshaping how developers interact with the kernel, yet he insists it remains just that—a tool, not a wholesale replacement for human programmers.
AI’s Unexpected Surge in Linux Development
Speaking alongside Dirk Hohndel, a fellow Linux kernel maintainer, Torvalds revealed a striking shift in development patterns. For two decades, the Linux kernel’s release process had maintained remarkable stability since adopting Git; however, this trend abruptly changed about six months ago as AI coding tools gained widespread traction. This marked a definitive turning point for the venerable operating system.
Torvalds noted a significant uptick in contributions, estimating a **20% increase in commits** for the last two releases compared to previous years. He initially misattributed this surge to excitement around the 7.0 release, but soon realized the true catalyst was the maturation of AI tools, which had become “good enough for a lot of people.” This led to a definite rise in development across nearly all fronts for the kernel.
Both Torvalds and Hohndel agreed that AI tools effectively lower the barrier to entry for new contributors, doing “a big chunk of the work.” However, Torvalds emphasized that the most profound impact was not purely technical, but rather **social**, forcing communities to adapt their established workflows. The traditional pain points in Linux, and likely most projects, have always been about human interaction and process changes, rather than the code itself.
Navigating AI-Driven Security and Disclosure
One of the most immediate “pain points” has emerged within the Linux kernel security mailing list, a traditionally small and confidential channel. Torvalds recounted how this critical list was recently “overrun by duplicate reports” generated by AI. Many developers, upon finding a potential bug with AI, instinctively report it to the security list, leading to a flood of redundant information that overwhelms the small team.
This influx forced the few maintainers to spend valuable time forwarding reports rather than addressing genuine threats. To mitigate this, Torvalds announced new, blunt **AI security disclosure guidelines**: if a security bug is found using AI, it should be considered public knowledge. The rationale is simple: if one person found it with AI, countless others have likely done the same, making true confidentiality impossible.
Despite this push for public transparency on AI-found bugs, Torvalds strongly urged researchers not to publish working exploits for actual security issues. He emphasized that researchers should avoid “crows about it publicly” and potentially enable malicious actors. This advice reflects a broader shift where AI-accelerated analysis means bugs are identified and publicized much faster, often within hours of a fix being deployed, leaving less time for quiet patching.
Torvalds firmly rejected the notion of closing source as a solution to AI-driven security concerns. He warned that AI can just as easily reverse-engineer closed-source systems to find vulnerabilities. In fact, he argued, closed-source might be “even worse” because AI cannot then assist in fixing the problems it exposes. This trend is evident beyond Linux, with Microsoft patching **1,139 CVEs in 2025**, a number experts anticipate will rise further in 2026 as AI tools become more adept at discovering security flaws.
The Core of the “Love-Hate” Relationship
Despite the frustrations, Torvalds reiterated his genuine appreciation for AI’s technical prowess, stating, “I actually really like it from a technical angle. I love the tools. I find it very useful and interesting.” He views AI-discovered bugs as “short-term pain” leading to “long-term benefits,” ultimately resulting in better, more robust software. After all, the real problem lies in the bugs that remain undiscovered.
However, the “hate” aspect centers on the **social strain** AI places on open-source communities. The sheer volume of AI-generated bug reports, especially for smaller projects with limited maintainers, can lead to significant burnout. When users submit drive-by reports and then fail to respond to requests for more information, it exacerbates the problem, creating “social choke points.”
Torvalds, as a top-level maintainer, highlighted that his primary role isn’t coding, but rather “working with people.” He firmly stated, “I do not use AI to work with people. Thank you. And I should suggest you don’t do that either.” This underscores a critical boundary: while AI assists with code, human interaction, community management, and mentorship remain fundamentally human tasks that no algorithm can replicate.
AI: A Productivity Multiplier, Not a Programmer Replacement
Addressing widespread concerns about AI replacing programmers, Torvalds pushed back strongly against exaggerated claims, stating, “When I see people saying, ‘hey, 99% of our code is written by AI,’ I literally get angry.” He compared AI’s role to that of compilers, which revolutionized productivity by a factor of 1000. He estimates AI will boost productivity by a factor of **10**, positioning it as another tool in a long line of technological advancements from hand-entered machine code to assemblers and compilers.
“AI is great, but AI is not changing programming,” Torvalds asserted, emphasizing that the fundamentals remain constant. He envisions AI generating code that compilers then process, much like previous tools layered upon each other, making it “revolutionary in the same sense that we’ve seen revolutions before.” This continuous evolution empowers developers without fundamentally altering their core responsibilities.
Crucially, Torvalds advised aspiring developers to maintain a deep understanding of the underlying systems, even when using AI. He shared his own practice of reviewing AI-generated code and even the assembly language, ensuring he comprehends the final output. For any serious, long-lived system, understanding “not just your prompts, but you need to understand the end result too,” is paramount for long-term maintenance and problem-solving.
Ultimately, Torvalds believes that AI, much like open source itself, is a powerful mechanism for managing software complexity. It is another valuable addition to the programmer’s toolkit, enhancing capability without diminishing the essential need for human judgment, community collaboration, and fundamental technical knowledge. Software remains incredibly complex, and open source, now augmented by AI tools, remains the best way to manage that complexity.
Source: ZDNet – AI
