On November 13, 2026, Polish developer Jakub T. Jankiewicz published a short article of roughly 1,000 words on his personal blog, arguing that all AI-generated code should be forced to be open source. The article received 968 votes and 539 comments on Hacker News.
This touches on a reality that companies avoid: when AI assistants generate hundreds of millions of lines of code every day, their legal ownership remains undetermined.
Two Blades: Training Data and the Copyright Gap
Jankiewicz's argument has two lines of logic.
The first concerns training data. He points out that mainstream AI coding assistants are trained on massive amounts of open-source code, including Stack Overflow content under attribution-style Creative Commons licenses. Models trained on these collective contributions generate closed-source commercial code, which he calls "license laundering."
The second concerns the copyright subject. Part 2 of the U.S. Copyright Office's AI report, released in January 2025, makes clear that works generated entirely by AI with no substantial human creative input are not protected by copyright. The conclusion traces back to the "monkey selfie" case—photos taken by a macaque in 2011 were ruled unprotected by both the court and the Copyright Office because the photographer was not human, and were placed in the public domain. Wikimedia Commons now requires all purely AI-generated images to be marked as public domain.
Combining these two lines, Jankiewicz concludes that developers and companies should proactively release AI-generated code under open-source licenses, or contribute it to the public domain under CC0.
Data Arrived First: The Problem Is Already Happening
Black Duck's 2026 Open Source Security and Risk Analysis report shows that among commercial codebases audited that year, 68% had open-source license conflicts, up 12 percentage points from 56% the previous year—the largest single-year increase on record. Some codebases had 2,675 distinct license conflicts. The report notes that AI coding assistants are a major driver: code snippets generated by models may carry obligations from strong copyleft licenses such as GPL or AGPL, but no license information is attached to the output.
On the governance gap, Black Duck's survey shows that 76% of enterprises check AI-generated code for security risks, but only 54% assess intellectual property and license risks, and 24% conduct a comprehensive three-in-one assessment covering security, quality, and IP.
The Real Divide on Hacker News
Jankiewicz's article manages to offend two highly overlapping groups at once.
The strongest analogy in opposition comes from a top comment: AI is a tool, like a compiler. A programmer's architectural decisions, prompt engineering, and code review cannot be dismissed because the last few lines were output by AI.
That analogy was rebutted: if GPL-licensed source code is fed into a compiler, the resulting binary is still bound by the GPL; why should output be exempt after being fed into an LLM for training? Another argument notes that not all developers can afford a $200-per-month subscription, and Anthropic does not support every country and region.
The debate exposes this: when AI code generation far outpaces human review, the premise that "developers bear substantive responsibility for the output" is already hard to sustain.
The Law Is Catching Up, but Still Far Behind
The U.S. Court of Appeals for the Ninth Circuit heard Doe v. GitHub in February 2026 on whether GitHub Copilot and other AI coding assistants infringe open-source licenses; as of publication, no ruling has been issued. The key question in the case is whether highly similar AI-generated code constitutes "verbatim copying" of training data.
UK law gives limited copyright protection to computer-generated works, while the U.S. holds that such works are not copyrightable. Jankiewicz argues that proactively adopting CC0 or open-source licenses is the simplest path to cross-jurisdictional consistency before global legal frameworks converge.
Who Is Delaying the Problem
The biggest obstacle to "mandatory open-sourcing of AI-generated code" is the structure of interests. Large tech companies offering AI coding assistants rely on a business model in which users generate and retain closed-source code. Requiring outputs to be open source would weaken enterprise customers' competitive moats.
A large number of small and medium-sized software companies are already using AI to generate code, yet they lack the capacity to review license compliance line by line and the resources to handle potential litigation. The 68% license conflict rate in Black Duck's report mainly describes the current situation of these companies.
Independent Judgment
Jankiewicz's argument is morally sound but operationally ambiguous—"what proportion must be human-written to count as a human work" has no enforceable definition in any legal system.
The more important conclusion is that mandatory open-sourcing may not be the only way out, but "doing nothing" is already producing consequences. A 68% license conflict rate and only 54% of enterprises assessing IP risks in AI code show that the industry's current default is to accumulate legal exposure through collective silence.
© 2026 Winzheng.com 赢政天下 | 转载请注明来源并附原文链接