AI-Generated Code Can Evade Copyleft Licensing, chardet Dispute Reveals
The maintainer of Python's chardet library has used Anthropic's Claude to create a clean-room rewrite — with under 1.3% structural similarity to the original — and relicensed it from LGPL to MIT. The dispute exposes how AI-assisted development could undermine copyleft licensing.
The Dispute
- chardet 7.0: Complete rewrite using Claude, now MIT-licensed
- Original creator Mark Pilgrim: Argued Blanchard had "ample exposure" to original code and no right to change license
- Maintainer Dan Blanchard: JPlag analysis shows max 1.3% similarity, "nothing was carried forward"
- Claude listed as contributor on the project
Why It Matters
If AI can produce functionally equivalent code with negligible structural similarity to the original, copyleft licenses become unenforceable. The LGPL requires derivative works to remain under the same terms — but if an AI generates new code from a natural language description of the same functionality, is it a derivative work?
The Practical Outcome
- 48x speed increase in character encoding detection
- ~130 million downloads per month
- Path cleared for inclusion in Python standard library
- Claude completed the rewrite in roughly five days
Bruce Perens' Take
Open source pioneer Bruce Perens argues this shows how AI will kill software licensing. The fundamental question: if AI ingested copyleft code during training, does its output infringe? If the output is structurally distinct, current legal frameworks have no answer.
Armin Ronacher's Observation
Flask creator Armin Ronacher noted: "Software licensing and the laws around it haven't been tested a lot in this new world of AI-assisted development."
Source: The Register