The dispute over chardet showed how AI is changing the rules of open source and code licensing

The story around the chardet library has become one of the most notable disputes about AI in open source. The maintainer rewrote the project with the help of an AI agent, changed the license, and essentially proposed to the community to decide whether to consider it a new product or a controversial duplicate of the old one.

How the dispute started

At the center of the story is developer Dan Blanchard, who has maintained chardet for over 12 years. In early March, he released a new version of the library, claiming it was written from scratch and therefore can be distributed first under MIT, and then under 0BSD instead of the previous LGPL. According to him, the goal was practical: to speed up the code, improve accuracy, and remove the license restriction that prevented chardet from being included in Python's standard library.

Blanchard described the process as a "clean room." First, an agent framework helped him gather specifications: compatibility with the public API, work on CPython and PyPy, emphasis on multi-threading and performance. Then these instructions went to Claude, which generated a new implementation.

Verification through JPlag showed just over 1% matches with the latest LGPL version, and the maintainer called this evidence in favor of the legitimacy of the approach. But even such a transparent scheme did not placate the community: dozens of contributors opposed the license change.

Where the line is drawn

The main question sounds simple: can code be considered "clean" if the person formulating the specification has worked with the original project for many years and knows its structure well? This is where critics see the weak point in the entire construction. The original author of the library, Mark Pilgrim, publicly objected that generative tools don't give the maintainer any additional rights to re-license. For his supporters, the problem is not with AI itself, but with an attempt to circumvent copyleft through a formal replacement of the implementation.

"Using a sophisticated code generator doesn't give any additional rights."

The legal part of the dispute is also far from clear. The article notes that Russia is discussing a bill under which AI-generated content could receive protection as a copyrighted work if recognized as original. In the USA, by contrast, in March 2026 a stricter approach was confirmed: work without meaningful human contribution does not receive copyright protection. This creates a paradox: if the new version of the library was indeed almost entirely created by an agent, then not only the license transfer becomes controversial, but also the question of whether such code has a full-fledged copyright holder at all.

What the practice changes

Blanchard's supporters have their own logic. They remind us that the clean room method is older than the current wave of AI: compatible systems were rewritten this way back in the 1980s, and today models merely drastically reduce the cost and time of such work. In this view, the chardet story is not an attack on open source, but a new engineering technique. Its supporters believe that a module rewritten from scratch more quickly is sometimes more useful than dependence on an old license, a single maintainer, or an abandoned repository.

• Can quickly obtain a compatible replacement of a critical library
• Easier to adapt a project to a permissive license and corporate use
• Reduces the risk of sabotage or compromise of a package in the supply chain
• The community gets a cheaper way to fork and create competing implementations

From this also arise new risks. If large companies begin to massively "rebuild" other people's open products with the help of agents, disputes about licenses will quickly turn into a struggle for ecosystems and audience. The article reminds us of the OpenTofu fork after Terraform's license change, as well as fresh examples where cloud companies reproduce other people's tools in a matter of days. In such an environment, the deciding factor remains not only the law, but also community trust: developers may simply not switch to the new branch, even if it is formally flawless.

What this means

The dispute around chardet shows that AI in open source changes not only the speed of development, but also the logic of code ownership itself. Now the community will have to renegotiate where architectural inspiration ends and the transfer of someone else's project under a new name and license begins.