Chardet dispute shows how AI will kill software licensing, argues Bruce Perens

Steven

Chardet dispute shows how AI will kill software licensing, argues Bruce Perens

Summary

Maintainer Dan Blanchard rewrote the popular Python library chardet with the help of Anthropic’s Claude and relicensed it from the LGPL to the permissive MIT licence. He says the new code is a “clean-room” AI-assisted reimplementation with under 1.3% similarity to prior releases and a 48x speed improvement, paving a possible path to Python’s standard library.

The change has provoked a dispute: an individual claiming to be original author Mark Pilgrim argues the LGPL requires derivatives to remain under the same licence, while others — including Armin Ronacher — note AI makes trivial rewrites easy. The Free Software Foundation warns that LLMs trained on licensed code undermine copyleft’s intent. Bruce Perens warns this is existential for software economics: if AI can reimplement code easily, both proprietary and copyleft models face large disruption.

Key Points

  1. Chardet 7.0 was reimplemented with AI (Anthropic’s Claude) and relicensed from LGPL to MIT by maintainer Dan Blanchard.
  2. Blanchard reports a 48x speed increase and claims near-zero similarity to prior code when analysed with JPlag.
  3. A dispute arose over whether the LGPL’s copyleft obligations can be sidestepped via AI-assisted rewrites; an alleged original author contends the licence still applies.
  4. FSF and other copyleft defenders argue LLM training on licensed code undermines copyleft and user freedoms.
  5. Bruce Perens warns the economics and legal frameworks around both proprietary and open-source software may be fundamentally altered by AI’s ability to reproduce code.
  6. Broader legal questions remain unsettled — e.g. how courts treat AI-generated or AI-assisted code — with recent cases like Thaler v. Perlmutter illustrating ongoing uncertainty.

Context and relevance

This isn’t just a library quarrel — it’s a live example of how large language models can change the rules that underpin software distribution and business models. For maintainers, lawyers and product teams this raises urgent questions about licence enforcement, provenance, and the value of code-as-asset. For the wider industry it suggests more niche and hardware-driven offerings may emerge while traditional licensing strategies weaken.

Author’s take (Punchy)

Big deal? Big deal. This dispute is a bright, noisy spotlight on a new reality: when AI can reproduce working code quickly and cheaply, licences that relied on friction and copyright enforcement start to lose teeth. Read the detail if you care about the future of software business models and the legal scaffolding that supports them.

Why should I read this?

Because someone just used an LLM to rewrite a widely used library, switched the licence, and kicked off a debate that could change how software is built, shared and monetised. It’s short, it’s topical, and it saves you the trouble of trawling forums to understand what this means for your projects.

Source

Source: https://www.theregister.com/2026/03/06/ai_kills_software_licensing/