Copyleft

Copyleft

Copyleft is a licensing condition for software and other works: you may freely use, modify, and redistribute them – but only under the same terms. This keeps the source code open in all later versions as well.

Whoever writes a program automatically owns the copyright to it. So they get to decide who is allowed to copy, modify, and redistribute it. Normally, companies use this right to exclude others. Copyleft turns the idea on its head: the author expressly permits everything – copying, modifying, selling – but ties this to one condition. Whoever redistributes the modified program must pass it on under exactly the same rules and must also supply the program’s blueprint, the so-called source code. The name is a play on words: the English “copyright” becomes “copyleft” – the opposite of it.

Why freedom without a condition doesn’t hold

Without copyleft, anyone could take an open piece of software, make a small change to it, and sell the new version as closed software. The original developers' work would then end up in a product that no one else is allowed to keep working on. That is precisely what the programmer Richard Stallman wanted to prevent back in the 1980s. To that end, he drafted the GNU General Public License, GPL for short, still the best-known copyleft license today.

The effect is a kind of chain reaction. Every improvement someone publishes inevitably flows back into the shared pool. Over the decades, this has produced a huge amount of freely usable software. The Linux kernel, the core of almost all servers and all Android phones, is licensed under the GPL.

For companies, this is also a risk. Anyone who incorporates copyleft code into their own product may, under certain circumstances, have to disclose their entire own source code. Large companies therefore employ lawyers to check which license is attached to which snippet of code. In the industry, this strict effect is sometimes mockingly called a “viral license”.

The condition that gets passed down

Technically, copyleft is not a law of its own but a contract. The license uses ordinary copyright as a lever. Without the author’s permission, one would not be allowed to use the code at all. That permission is only granted bundled together with the obligation to pass it on. Whoever fails to comply loses the right of use and thereby infringes copyright.

A distinction is made between strong and weak copyleft. Under strong copyleft, such as the GPL, the obligation applies to the entire program into which the code is incorporated. Under weak copyleft, such as the LGPL, it applies only to the modified program library itself. The rest of the application may remain closed. A third group, the permissive licenses such as MIT or Apache, dispenses with this obligation altogether: you only have to credit the author.

A common misconception: copyleft does not mean that software must be free of charge. You are allowed to sell a GPL program. The buyer then receives the source code, though, and may redistribute it themselves – which makes high prices difficult in practice.

Copyleft in AI models and open datasets

In everyday life, anyone who reads Wikipedia encounters copyleft. Its texts are published under a Creative Commons license with a copyleft condition, abbreviated “Share Alike”. Anyone who takes Wikipedia text and expands it must make the result freely accessible again as well. The Firefox browser, the Linux operating system, and many graphics programs are likewise built on copyleft licenses.

In the AI industry, the question is new and unresolved. If a language model has learned from copyleft code and outputs similar code – does the condition then also apply to that output? Several lawsuits are underway over exactly this question, including one against the programming assistant GitHub Copilot. So far, there is no definitive answer.

At the same time, licenses for the models themselves are emerging. Some open models do allow downloading the weights but prohibit certain uses or large-scale commercial use. Purist copyleft experts do not consider this genuine copyleft, since it restricts uses rather than merely regulating redistribution. So when news reports speak of “open source” in connection with AI models, it is worth taking a closer look at the specific license.

Subscribe free. Unsubscribe the second it sucks.

High-signal news across AI, business, UX, and tech. Every morning.