GPL

Whoever writes a program automatically owns the rights to it. Others may therefore not simply copy or modify it. With a license — that is, a fixed text containing permissions and conditions — the author can pass on these rights. The GPL is such a license text, and indeed one of the best-known in the world. It allows anyone to use, examine, modify, and redistribute the program. In return, it demands something: whoever redistributes the program must grant recipients the same freedoms and provide the readable program text along with it. The abbreviation stands for General Public License.

Why voluntary sharing became a legal document

Until the 1980s, programmers often shared their code loosely among themselves. Then companies began selling software only in machine-readable form. The readable program text, the source code, remained secret. This meant no one could fix bugs themselves or adapt the program to their own needs.

The programmer Richard Stallman published the first GPL in 1989 as a counter-model. Its clever trick is an inversion: it uses copyright not to forbid copying, but to enforce openness. This principle is called copyleft. A company is thus free to use GPL software and even sell it. However, it may not turn it into a closed product whose source code no one ever sees.

This is the key difference from other open licenses such as MIT or Apache. These also permit almost everything but do not require disclosure upon redistribution. Whoever takes MIT-licensed code may build it into a secret product. With the GPL, this is not possible. That is why it is called a 'strong' or 'contagious' license.

What copyleft specifically requires

The GPL’s obligations only kick in upon distribution to third parties. Whoever merely modifies a program internally and uses it in-house does not need to publish anything at all. But as soon as the modified program is delivered to customers, the following applies: the complete source code must be provided or made accessible. In addition, the recipient must receive the same rights that one had oneself.

If GPL code is closely combined with one’s own code, the result is usually considered a single work. In that case, one’s own portion also falls under the GPL. This is exactly what companies are wary of, which is why large firms scrutinize very carefully which licenses are contained in their products. There are court rulings for violations: in Germany, among others, the developer Harald Welte has repeatedly and successfully sued router manufacturers who shipped Linux code without the source text.

There are several versions. Version 2 from 1991 is still used in Linux today. Version 3 from 2007 additionally closes loopholes, such as devices that do provide the source code but technically prevent modified software from starting. There is also the LGPL, a weakened variant for program components that may also be incorporated into closed software.

GPL software in everyday life and in the AI debate

One encounters the GPL every day without noticing. The kernel of the Linux operating system is licensed under GPLv2 and runs on most of the internet’s servers. GPL code is also often found in routers, televisions, and cars. That is why many devices come with a slip of paper containing license notices that hardly anyone reads.

In the news, the GPL usually comes up in connection with disputes. One example is the debate around coding assistants such as GitHub Copilot: these AI systems have learned from, among other things, GPL code and sometimes suggest lines that match almost verbatim. Whether the licensing obligations travel along with this is, to this day, not conclusively settled legally.

A common misconception is also that 'GPL' and 'open source' are the same as 'free of charge.' The GPL does not expressly prohibit selling. Nor does the term 'open-source AI model' usually refer to the GPL: models such as Llama are released under their own licenses with restrictions that contain no copyleft.