GPLv3

GPLv3

The GPLv3 is a license text from 2007 that sets out the conditions under which a computer program, together with its source code, may be redistributed. Its core rule: anyone who modifies and distributes the program must also disclose the modified source code and pass on the same rights.

Every computer program is written by humans in a human-readable form, the so-called source code. Whoever has this source code can understand and modify the program. Normally, a company releases only the finished program and keeps the source code to itself. The GPLv3 is a license text that prescribes exactly the opposite. It allows anyone to use, copy, modify, and resell the program. In return, it demands one condition: whoever passes on a modified version to others must also supply its source code and grant the same freedoms. The abbreviation stands for Version 3 of the GNU General Public License, published in 2007 by the Free Software Foundation.

The contagion effect and why companies fear it

The GPLv3 is considered a so-called copyleft license. Copyleft is a play on words on copyright and means: freedom passes on down the chain. If someone builds GPL code into their own program and sells that program, the entire result must again be placed under the GPL. This is why people half-jokingly speak of a contagious license.

For companies, this is a serious decision. A software house that wants to keep its product secret must not copy any GPLv3 code into it. Many large companies therefore have internal rules about which licenses their developers may use. Alternatives such as the MIT License or the Apache License do not require copyleft. With them, one may also build closed products, only having to credit the authors.

Conversely, it is precisely this strictness that protects the community. The Linux kernel, the heart of Android and almost all internet servers, is under the predecessor version GPLv2. Without copyleft, every company could have built its own, secret variant. The shared commons would have fallen apart.

What Version 3 adds compared to Version 2

Version 3 came about because two gaps in Version 2 became apparent. The first concerns devices that do contain free software but are technically locked down. A video recorder manufacturer dutifully supplied the source code, yet the device would only run software with a certain digital signature. So one was allowed to modify it, but not to use it. The GPLv3 forbids this and requires that the necessary installation information be provided as well. This clause is unofficially named Anti-Tivoization, after the manufacturer.

The second innovation concerns patents. A patent is a state-granted monopoly on a technical idea. Whoever distributes software under the GPLv3 automatically grants all recipients a permission for their own patents covering that code. They cannot later sue the users. This was meant to prevent someone from giving away code and then afterward demanding license fees for the ideas contained within it.

A common misconception: the GPLv3 does not prohibit selling. One may charge money for a GPL program, as much as one likes. It’s just that the buyer must receive the source code and be allowed to pass it on in turn. Likewise, the license does not apply as long as one uses a modified program only for oneself. Only distribution to third parties triggers the obligations.

Where GPLv3 turns up in the news and on your own computer

Under GPLv3 are, among others, the graphics program GIMP, parts of the video software VLC, and many components of the GNU system that is built into every Linux distribution. On platforms like GitHub, where programmers publish their code, the license is stored in a file called LICENSE. Anyone downloading a project should check there first.

In the AI debate, the GPLv3 plays an indirect but growing role. Models such as Llama or Mistral are often described as open, but are actually under their own licenses with additional conditions. Experts dispute whether such models are even free in the sense of the GPL. One point of contention is what corresponds to source code in an AI model: the program code, the training data, or the trained numerical weights.

This also becomes practically relevant for programming assistants. Such tools were partly trained on GPL code and sometimes suggest lines of code derived from it. Whether this gives rise to license obligations is legally unresolved. This is precisely why a license text more than 15 years old regularly ends up in current tech news.

Subscribe free. Unsubscribe the second it sucks.

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