Open source

Open source is software whose source code anyone can read, change and distribute. Openness is defined by the license: according to the Open Source Initiative definition, “Open source doesn’t just mean access to the source code.” The license must allow the program to be freely distributed, modified, and distributed in modified versions.¹

The movement began in 1983 with the free GNU system project, and the term “open source” appeared in 1998.²³ Today open source is found in almost every program, and its value to the economy is estimated in the trillions of dollars.⁴⁵ With the arrival of AI it has also become raw material: the models that write code were trained on open source, while the projects themselves suffer from AI crawlers and a flood of AI-generated changes.

History

• 27.09.1983Richard Stallman announces that he is starting to write a complete Unix-compatible system, GNU, and will “give it away free to everyone who can use it”.² ¶ ckb4b47nh
• 25.08.1991Linus Torvalds writes in the comp.os.minix group that he is making a free operating system for 386 computers: “just a hobby, won’t be big and professional like gnu”. It grew into Linux.⁶ ¶ c2mp68k6u
• 03.02.1998At a meeting in Palo Alto, shortly after Netscape decided to open the source code of its browser, the term “open source”, proposed by Christine Peterson, is born; the Open Source Initiative is founded at the end of February.³ ¶ ct64k8kz9
• 10.04.2008GitHub launches, the hosting service that became the main home of open source.⁷ ¶ c7deb3mcs
• 29.06.2021GitHub unveils Copilot, based on the OpenAI Codex model, which “was trained on a data set that includes a much larger concentration of public source code”.⁸ ¶ cav3fkjdg
• 03.11.2022Programmers file a class action against GitHub, Microsoft and OpenAI over Copilot; according to the plaintiffs, it is “the first class-action case in the US challenging the training and output of AI systems”.⁹ ¶ cwzsrimk7
• 28.10.2024The Open Source Initiative publishes the Open Source AI Definition 1.0: not only the weights must be open, but also “Sufficiently detailed information about the data used to train the system so that a skilled person can build a substantially equivalent system”.¹⁰ ¶ c6gytcsan
• 16.09.2026The US Court of Appeals for the Ninth Circuit upholds the dismissal of the DMCA claims against Copilot: in the court’s view, Copilot and Codex do not “remove or alter” copyright management information “from a copy of an existing protected work but instead create new works that never contained that information”.¹¹ ¶ c4c4ngdr8

Data

Incidents

A backdoor in xz, March 2024 ¶ cjfj5sjxe

On 29.03.2024 developer Andres Freund reported: “The upstream xz repository and the xz tarballs have been backdoored.” The backdoor in the xz/liblzma compression library allowed an ssh server to be compromised (CVE-2024-3094).¹⁴ The backdoored tarballs were created and signed by a contributor going by the name Jia Tan.¹⁵ Two years earlier, project maintainer Lasse Collin had written that “my ability to care has been fairly limited mostly due to longterm mental health issues”, and of Jia Tan: “perhaps he will have a bigger role in the future”.¹⁶

AI crawlers versus open-source infrastructure, 2025 ¶ ckrvdjbad

Drew DeVault, founder of the code hosting service SourceHut, wrote on 17.03.2025: “I have spent anywhere from 20-100% of my time in any given week mitigating hyper-aggressive LLM crawlers at scale”; the service suffers “dozens of brief outages per week”; and “If you think these crawlers respect robots.txt then you are several assumptions of good faith removed from reality.”¹⁷ The GNOME community put Anubis in front of its GitLab, a proxy that “uses proof-of-work challenge to block potential AI scrapers”.¹⁸

Wikimedia: bot traffic, April 2025 ¶ ciz668dck

The Wikimedia Foundation reported that since January 2024 the bandwidth used for downloading media files has grown by 50%, and the growth comes not from people but from automated programs: “at least 65% of this resource-consuming traffic we get for the website is coming from bots”, even though bots account for about 35% of page views.¹⁹

Criticism

The Open Source Initiative has twice stated that the license of the Llama models from Meta is not open: on 20.07.2023, that “the tech giant has created the misunderstanding that LLaMa 2 is “open source” – it is not.”, and on 18.02.2025, that “Llama 3.x is still not Open Source by any stretch of the imagination”, even though Meta keeps calling it that.²⁰²¹ ¶ ceqg7vn5e

Andreas Liesenfeld and Mark Dingemanse (Radboud University), after studying more than 45 generative AI systems, called this phenomenon open-washing: the term “open source” is used widely, but “many models are ‘open weight’ at best and many providers seek to evade scientific, legal and regulatory scrutiny by withholding information on training and fine-tuning data”.²² ¶ ce32w57y8

Evolution

Some major projects are shutting out AI-written code. In April 2024 the Gentoo Council ruled: “It is expressly forbidden to contribute to Gentoo any content that has been created with the assistance of Natural Language Processing artificial intelligence tools.” NetBSD says such code is “presumed to be tainted code” and does not accept it without written permission from the project leadership; QEMU will “DECLINE any contributions which are believed to include or derive from AI generated content”.²³²⁴²⁵ ¶ cefpkc7aa

The other side of the same process is a flood of unchecked AI-generated changes and reports that leaves maintainers unable to cope with the load: see maintainer crisis.