Five days. That is how long it took one developer to rewrite a Python library that gets 130 million downloads per month, change its license, and set off a firestorm about the future of open-source software.
On March 4, Dan Blanchard released chardet 7.0.0, a "ground-up, MIT-licensed rewrite" of the popular character encoding detection library. The original chardet was created by Mark Pilgrim in 2006 under the LGPL (Lesser General Public License), a copyleft license that requires derivative works to keep the same license. Blanchard used Anthropic's Claude Code to produce the rewrite, which he says resulted in a 48x speed improvement and shares less than 1.3% similarity with any prior version according to plagiarism detection tools.
Pilgrim, who had largely retired from public internet life since 2011, came back to open GitHub Issue #327 titled "No right to relicense this project." His argument: Blanchard had been maintaining chardet for over a decade, was deeply familiar with the original code, and adding "a fancy code generator" does not magically create a clean room implementation.
The "Clean Room" Problem
Traditional clean room engineering works like this: one team studies the original product and writes a specification, then a completely separate team builds a new implementation from that spec alone. The wall between the two teams is what makes it legally defensible.
Blanchard's process collapses both roles into one person. He says he started in an empty repository and explicitly instructed Claude not to reference LGPL or GPL code. But he spent a decade immersed in chardet's internals. At least one documented instance shows Claude Code referenced parts of the original codebase during the rewrite. Whether that matters legally is genuinely unclear, because no court has ruled on this exact scenario.
Simon Willison, co-creator of Django, said he is "personally leaning towards the idea that the rewrite is legitimate, but the arguments on both sides are entirely credible." That is about as honest a summary as you will find.
The Deeper Threat to Copyleft
The GPL family of licenses has always depended on one practical assumption: rewriting a large codebase from scratch is so expensive that nobody bothers. You either use the code and follow the license, or you write your own from zero. AI has shattered that cost barrier.
Bruce Perens, who wrote the original Open Source Definition, told The Register that "the fundamental assumptions behind software licensing were built for a world where humans wrote every line of code." His proposed fix is radical: abandon copyright-based licensing entirely and move to a contractual revenue-sharing model, similar to how ASCAP handles music royalties.
That might sound extreme, but the alternative is not great either. Companies that relied on AGPL or GPL to prevent proprietary competitors from cloning their work now face the real possibility of AI-generated replicas that owe nothing back to the original project. The term "slopfork" has already emerged to describe these AI-generated rewrites that reproduce behavior while shedding legal obligations.
Meanwhile, U.S. copyright law is sending mixed signals. On March 2, the Supreme Court declined to hear an appeal on AI-generated copyrights, reinforcing that AI-generated material needs human authorship to be copyrightable. That creates a paradox: if the AI-rewritten code is not copyrightable, then Blanchard's MIT license claim on it may not hold up either.
What This Means for Developers Using AI Coding Tools
For most developers using tools like Claude Code, Cursor, or GitHub Copilot for everyday coding, this is not an immediate crisis. Writing new functions, debugging, and refactoring your own code remain clearly fine.
The line gets blurry when you use AI to systematically rewrite an entire project you did not author, especially one with copyleft obligations. The chardet case is likely headed for legal action or, at minimum, will force package managers like PyPI to develop policies around AI-relicensed packages.
The Free Software Foundation has called for community research on these questions but has not issued a specific position on AI-assisted relicensing yet. Until courts or the FSF weigh in definitively, any developer using AI to rewrite copyleft code and slap a new license on it is operating in a legal gray zone. The 1.3% similarity score might look clean on paper, but "clean" and "legal" are not the same thing.