Open Source

Every program consists of text that humans have written: a long list of instructions telling the computer what to do. This text is called source code. With most commercial programs, it remains a company secret. You can use the program, but you can’t look inside it. Open source is the opposite: the source code is public, and a license explicitly permits reading, modifying, and redistributing it. Well-known examples are the operating system Linux, the browser Firefox, and the encryption technology behind almost every website.

Why half the world runs on someone else’s code

Open code saves an enormous amount of duplicate work. Nobody needs to reinvent a function that has already been written ten thousand times. That’s why companies build their products on open building blocks that others have maintained. Estimates suggest that almost all commercial programs today contain open-source components.

A second advantage is control. Because anyone can look inside, bugs and hidden functions are easier to find. Government agencies and hospitals therefore like to rely on open software: they don’t want to depend on a single company. If a provider raises prices or discontinues a product, you can keep maintaining the code yourself.

But this advantage is not automatic. “Anyone could look inside” doesn’t mean someone actually does. In 2014, a widely used encryption library harbored a serious bug for years, even though the code was open. A project is only maintained if enough people have the time for it.

Licenses, forks, and who gets a say

The heart of open source is not the code, but the license. It is a legal text that specifies what users are allowed to do. Permissive licenses allow almost anything, including incorporation into paid products. Stricter licenses require that any modified version remain open as well. Anyone who fails to comply violates copyright law.

Development usually takes place on public platforms like GitHub. That’s where the code lives, where bugs are reported, and where outsiders propose changes. A small team of maintainers decides what gets adopted. If a group doesn’t like the direction, it can split off its own copy and continue developing it separately. This is called a fork.

It’s important to distinguish this from “free of charge.” Many apps are free, but their code remains secret. Conversely, you are also allowed to sell open-source software. Large companies make their money by operating, maintaining, and offering support for open software.

The dispute over open AI models

In AI news, the term comes up almost daily. Models like Meta's Llama, Mistral, or DeepSeek are marketed as open. You can download them and run them on your own machines instead of sending requests to someone else’s server. For companies with sensitive data, that’s a strong argument.

Experts disagree, however, on whether this is really open source. Usually, only the finished numerical values of the trained model, the so-called weights, are published. The training data and the exact training process often remain secret. So you get the cake, not the recipe. Some licenses also prohibit certain uses or large commercial providers, which contradicts classic open-source rules. “Open weights” would be the more accurate term.

Economically, this is a question of power. Open models drive down prices for closed services and shift market shares. Laws are tied to this as well: the EU’s AI legal framework provides for partly different obligations for freely available models than for closed ones. Anyone reading reports about “open-source AI” should therefore check exactly what has been disclosed.