What Happened
Dan Blanchard, maintainer of the Python chardet library, released version 7.0.0 under the MIT license on March 6, 2026 - breaking from the previous LGPL license. He used Anthropic's Claude to create what he calls a clean-room reimplementation, achieving a 48x speed improvement in five days of work.
The numbers behind this are significant. Chardet gets 130 million monthly downloads. Blanchard's plagiarism detection analysis shows version 7.0.0 has less than 1.3% structural similarity to prior versions. He says the goal was to get chardet into Python's standard library while improving performance - work he does as an unpaid volunteer.
An individual claiming to be original creator Mark Pilgrim pushed back, arguing that modified licensed code must stay under the same LGPL license and that using AI doesn't grant extra rights beyond copyright restrictions.
The debate drew in major figures. Flask creator Armin Ronacher welcomed the change, noting that copyleft code "depends on copyrights and friction" but code in the open can now be "trivially rewritten." Free Software Foundation's Zoë Kooyman cautioned against calling any LLM implementation "clean" when models ingest source code during training.
Bruce Perens, co-founder of the Open Source Initiative, went furthest: "The entire economics of software development are dead, gone, over, kaput!" He demonstrated creating an SRE platform by having AI examine competitor documentation and generate working code immediately.
Why It Matters
This isn't a theoretical debate. It's a working example of someone using an AI tool to rewrite a widely-used library, strip its copyleft license, and ship it as MIT. Whether or not this specific case holds up legally, the technique is now proven and repeatable.
For anyone who relies on open-source tools daily, the implications are direct. Copyleft licenses like GPL and LGPL exist to ensure improvements stay open. If AI can produce functionally equivalent code with less than 1.3% similarity, the legal mechanism that enforces "share alike" may not work anymore.
This also matters for AI tool users on a practical level. If you're using Claude, ChatGPT, or Cursor to write code, you're potentially generating output derived from copyleft-licensed training data. The legal status of that output remains unresolved, and this dispute is pushing toward a real test case.
Our Take
Perens is probably right that this changes software licensing permanently, even if his "everything is dead" framing is dramatic. The core problem is real: copyleft enforcement depends on proving derivation, and AI makes derivation nearly impossible to prove when the output is structurally different from the input.
But the FSF's point also stands. Calling an AI rewrite "clean-room" is a stretch when the model trained on the original code. Clean-room implementations traditionally require strict information barriers - the implementer never sees the original. An LLM that ingested the source during training doesn't meet that standard in any traditional sense.
For practitioners, the immediate takeaway: be aware of the licensing status of AI-generated code in your projects. If you're using AI tools to rewrite or replace existing libraries, you're stepping into legally untested territory. The chardet case will likely become a reference point for how courts and the open-source community handle this going forward.
The bigger question is whether new licensing models will emerge that account for AI-mediated copying. The current frameworks were built for a world where humans wrote code. That world is ending faster than the legal system can adapt.