AI Rewrites Copyleft Code and Changes License — Open Source Community Erupts

What Actually Happened

Here is the scenario that is keeping open source lawyers awake at night. A developer takes an LGPL-licensed project, feeds its codebase through an AI model, and publishes the result under an MIT license. The original author cries foul. The community picks sides. And the entire legal framework that has governed open source software for two decades finds itself in uncharted territory.

That is exactly what played out with a project covered by YouTube channel SavvyNik in a video that has racked up nearly 40,000 views. The details are specific — a maintainer used an AI / LLM rewrite to effectively strip the copyleft restrictions from LGPL code and relicense it as MIT. But the implications are anything but narrow. What we are witnessing is the collapse of the assumption that code licenses are enforceable when the code itself can be regenerated from scratch by a machine that has already seen everything.

How the AI Rewrite Actually Works

This is not about copying files. Nobody is running cp -r on a Git repo and changing the LICENSE file. The mechanism is far more insidious. A developer points an AI agent — Claude, GPT, or whichever frontier model — at an open source repository and gives it a single instruction: rebuild this from scratch without copying any source files.

The model, having been trained on millions of public repositories including the target codebase, generates a functionally identical application using entirely different syntax. Every algorithm is reimplemented. Every variable is renamed. The architecture may shift. But the behavior — what the software actually does — is preserved perfectly.

This is an automated clean-room design, compressed from months of manual effort into minutes. Agent A studies the original to produce a specification. Agent B writes new code from that spec alone. The problem? Both agents run on the same neural network that memorised the original code during training. The separation is formal, not substantive.

The Legal Paradox Nobody Has Solved

The situation creates a legal knot that no court has fully untangled. Under the Abstraction-Filtration-Comparison test established in Computer Associates v. Altai, courts filter out elements dictated by efficiency or standard practice before comparing programs for infringement. When an AI rewrites code, it naturally optimises the structure — removing the original author's unique creative expression. The resulting code looks entirely different to automated comparison tools.

Here is where it gets worse. Following rulings like Thaler v. Perlmutter, courts have held that human authorship is a requirement for copyright protection. If a codebase is generated entirely by an AI agent, the output cannot be copyrighted. It is effectively public domain. So the developer who used AI to rewrite LGPL code as MIT has created something that:

• Cannot be proven to infringe the original copyright (different syntax)
• Cannot itself be copyrighted (no human authorship)
• Can be copied by anyone, including the original project's competitors

This is not a loophole. It is a crater.

The Community Reaction — Copyright Laundering or Fair Game?

The open source community has not taken this quietly. The term being thrown around is "copyright laundering" — using AI to strip the legal obligations from copyleft code while retaining its functional value. The original authors of the LGPL project in question called for a revert. The fork's maintainers argue that since no literal source code was copied, no license violation occurred.

Both sides have valid points. The original authors spent years building a community under LGPL terms that explicitly require derivative works to remain under the same license. The fork creators are technically correct that their codebase shares no literal file content with the original. The law has never dealt with functional replication at this fidelity.

This mirrors the broader pattern we have seen in open source over the past two years. HashiCorp's switch to BSL spawned OpenTofu. Redis's license change created Valkey. Elastic's move to SSPL produced OpenSearch. Each of those forks was a human-driven response to a licensing decision. This is different. This is AI doing the forking for you — automatically, instantly, and without human authorship.

What This Means: The End of Copyleft as We Know It

Let me be direct about this. If AI can functionally replicate any open source project from a prompt, the enforcement model of copyleft licenses is broken. The GPL, LGPL, and AGPL all rely on the assumption that if you want to use or modify the code, you must accept its license terms. That assumption only holds when copying the code is the only practical way to use it. AI removes that constraint.

Projects that want to protect themselves are already exploring alternatives. The Business Source License adds a time delay before code becomes fully open. Server Side Public License restricts cloud hosting. Some projects are moving to proprietary licenses with source availability. But none of these were designed to defend against AI-driven functional replication.

The real question is whether the open source community needs to redefine what "derivative work" means in the age of generative AI. If a model trained on your LGPL code can reproduce its functionality on demand, is every output a derivative work? Current case law says no. The community's instincts say yes. This tension is not going to resolve itself.

What Comes Next

We are going to see three things happen in the next twelve months. First, a test case — someone will sue over AI-generated functional replication of copyleft code, and a court will have to rule on whether the output constitutes a derivative work. Second, license changes — projects that rely on copyleft will either adopt source-available licenses or add explicit AI training and output restrictions to their terms. Third, model training transparency requirements will become a licensing issue, not just an ethical one.

For developers building on open source today, the practical advice is simple: assume that any code you publish in a public repository can be functionally replicated via AI within minutes. If you are relying on copyleft to protect your work, that protection is now probabilistic, not guaranteed. And if you are using AI to rebuild someone else's code, understand that you are operating in a legal grey zone where the community's verdict may be harsher than any court's.

The open source ecosystem has survived license wars, cloud provider exploitation, and corporate takeovers. This is a fundamentally different challenge. The law has not caught up. The tools have not caught up. And the community is fighting about it in real time. That is exactly where the most important conversations happen.

— Allan Ali, Sylt.ing