Say you download a language model. It takes a few moments. A file named like a serial number lands on your laptop, and a minute later one of the systems everyone is arguing about is running on ordinary hardware in front of you. No license server checked in, and nobody from sales called. The release page calls it open. So do the announcement and the press coverage.
The argument over that word has institutions, court cases, and decades of history. None of it shows up next to the download button. You can use the model before you know which freedoms you received.
An older argument
About twenty-five years ago the same word sat at the center of a fight over operating systems. Linux was moving off hobbyist machines and into the infrastructure of real companies, and the established software industry didn’t welcome it. The line most people remember came from Steve Ballmer, then Microsoft’s chief executive, who told the Chicago Sun-Times in 2001 that Linux was “a cancer that attaches itself in an intellectual property sense to everything it touches.”1
Ballmer was aiming at the license. Linux is distributed under the GNU General Public License, which says that if you distribute a modified version, you have to make your changes available under the same terms.2 To a company that made its money selling closed software, that requirement looked contagious. Whether Linux worked barely came into it. The complaint was about what the license allowed, what it required, and what it would do to the business of anyone who built on it.
To be fair to Microsoft, it changed its mind. By 2014 its next chief executive, Satya Nadella, was standing in front of a slide proclaiming that Microsoft loves Linux,3 and the company has since become one of the larger corporate contributors to open source.4 I quote Ballmer because his line shows how long people have argued over what a software license means for a business. He certainly wasn’t the only one who saw it his way.
That argument never ended. It moved from operating systems to web servers, then to databases, developer tools, and cloud platforms. Now it has reached AI models: the weights (the learned numbers that make up a trained model), the training data, the code, and everything around them. At every stop, the two sides used the same word and meant different things by it.
Why models make it harder
Companies call their AI models open source in ways that confuse even experienced practitioners, me included. One model permits commercial use but requires a separate license for organizations whose products exceed a specified user threshold at the release date.5 Another ships its weights but withholds the training data and the code that produced them. Another lets you run and modify the model but restricts what you may use it for, through an acceptable-use policy attached to the license.6 Each of these announcements says “open.” They don’t mean the same thing, and some of them sit a long way from what the word has meant in software for decades.
I don’t think most of this is bad faith. Releasing weights that anyone can download and run is a real contribution, and a company can have honest reasons to hold back data or limit certain uses: cost, privacy, copyright exposure, competition, safety. The trouble is that one word covers all of those choices, and the person pressing the download button usually can’t tell which one they got.
What open source is
Underneath the confusion is something well established. Open source is the practice of sharing software so that others can read it, change it, redistribute it, and build on it, under licenses that grant those rights. Some also require redistributors to preserve them in covered versions. It has been part of the software industry for decades, and just about every major corporation depends on it.7 It’s in the network standards and infrastructure that carry traffic, the operating systems under phones and servers, and much of the software that keeps everyday systems working.
Open source is a technical and institutional field, funded and supported by some of the largest corporations in the world.8 Companies take part at that scale because it works for them. Corporate support does not define the whole field. Volunteers, foundation staff, and maintainers working nights and weekends are as much a part of open source as any corporate sponsor.
It’s also plural by design. There’s no single open-source license, but several families of them. Copyleft licenses, associated with the Free Software Foundation and the GNU project, use the license’s own conditions to keep software free for everyone who receives it later.9 Permissive licenses such as MIT and BSD allow broad reuse with few conditions.10 Apache 2.0 is permissive and includes an express patent grant. The Mozilla Public License applies copyleft at the file level: when covered files are distributed, recipients must be able to get their source under the license.11 Creative Commons covers documents, data, and cultural work, which is a different problem from software and matters a great deal once training data enters the picture.12 Foundations such as the Linux Foundation, the Apache Software Foundation, and the Eclipse Foundation provide legal and organizational homes for projects.13 The Open Source Initiative, founded in 1998, maintains the Open Source Definition that the phrase rests on,14 and the Free Software Foundation, founded in 1985, has kept its own older, rights-based definition since 1986.15
Free, as in freedom
One word trips up newcomers more than any other. In open source, free usually means freedom, not price. The Free Software Foundation spells this out as four freedoms: to run the program, to study and change it, to redistribute copies, and to distribute your modified versions. In its words, think of “free” as in “free speech,” not as in “free beer.”16
Price is a separate question. A program can cost money and still grant all four freedoms, and another can cost nothing and grant none of them. A model you download without paying is free as in beer. Whether it’s free as in speech depends on the license and the policies attached to it, and with models, that’s often where the word open starts to come apart.
Five questions
This book works from a short list of questions. Whenever something is called open, I ask:
- What has been opened?
- To whom?
- Under what terms?
- With which freedoms?
- With what ability to continue without permission?
They look simple. For a typical software project in the late 1990s, you could answer most of them by reading the license file. A model release is harder because a model is several things at once. There are the weights, maybe the training data, maybe the training code, a license, possibly an acceptable-use policy, and often a hosted service. Any one of those layers can be open while the others are closed. The questions won’t settle that for you. They will show you which layer you’re looking at, and that’s usually where an honest answer starts.
What the answers are for
A release can fail an open-source definition and still give you useful freedoms. It can satisfy a license checklist and still leave you dependent on the publisher. The questions help separate those judgments, so you can decide which freedoms your work requires and recognize when a claim of openness overstates the release.
Asking those questions well takes some vocabulary: which licenses exist, who wrote them, and what each side was trying to protect. Chapter 2 traces where that vocabulary came from, starting with a popular web server that had lost its maintainer.
Notes
Dave Newbart, “Microsoft CEO takes launch break with the Sun-Times,” Chicago Sun-Times, 1 June 2001, archived by the Internet Archive at https://web.archive.org/web/20010606035140/http://suntimes.com:80/output/tech/cst-fin-micro01.html (Ballmer’s full answer: “Linux is a cancer that attaches itself in an intellectual property sense to everything it touches. That’s the way that the license works.”); The Register, “Ballmer: ‘Linux is a cancer,’” The Register, 2 June 2001, https://www.theregister.com/2001/06/02/ballmer_linux_is_a_cancer/. ↩
Linux kernel documentation, “Linux kernel licensing rules,” kernel.org, https://www.kernel.org/doc/html/latest/process/license-rules.html (the kernel is provided under GPL-2.0 only); Free Software Foundation, “GNU General Public License, version 2,” section 2(b), https://www.gnu.org/licenses/old-licenses/gpl-2.0.html. ↩
Microsoft Windows Server Blog, “Microsoft Loves Linux,” Microsoft, 6 May 2015, https://www.microsoft.com/en-us/windows-server/blog/2015/05/06/microsoft-loves-linux (describes Nadella’s briefing slide, which read “Microsoft ♥ Linux”); Ars Technica, “Microsoft ‘loves Linux’ as it makes Azure bigger, better,” Ars Technica, October 2014, https://arstechnica.com/information-technology/2014/10/microsoft-loves-linux-as-it-makes-azure-bigger-better. ↩
Microsoft Azure Blog, “Microsoft’s open source journey: From 20,000 lines of Linux code to AI at global scale,” Microsoft, 2025, https://azure.microsoft.com/en-us/blog/microsofts-open-source-journey-from-20000-lines-of-linux-code-to-ai-at-global-scale. Microsoft’s own account; it cites CNCF contribution statistics and Microsoft’s 2009 kernel contribution. ↩
Meta, “Llama 3.1 Community License Agreement,” section 2, “Additional Commercial Terms,” 23 July 2024, https://www.llama.com/llama3_1/license/. Licensees whose products exceeded 700 million monthly active users must request a separate license from Meta. ↩
Meta, “Llama 3.1 Acceptable Use Policy,” https://www.llama.com/llama3_1/use-policy/. ↩
Black Duck, “2026 Open Source Security and Risk Analysis Report,” March 2026, https://www.blackduck.com/resources/analyst-reports/open-source-security-risk-analysis.html. 98 percent of the commercial codebases Black Duck audited contained open source components. ↩
Linux Foundation Research, “2024 Open Source Software Funding Report,” Linux Foundation, 2024, https://www.linuxfoundation.org/research/open-source-funding-2024. Surveyed organizations invest an estimated $7.7 billion a year in open source, mostly as paid labor; the survey self-reports and targets organizations likely to contribute. ↩
Free Software Foundation, “GNU General Public License, version 3,” https://www.gnu.org/licenses/gpl-3.0.html. ↩
Open Source Initiative, “The MIT License,” https://opensource.org/license/mit; Open Source Initiative, “The 3-Clause BSD License,” https://opensource.org/license/bsd-3-clause. ↩
Apache Software Foundation, “Apache License, Version 2.0,” section 3, “Grant of Patent License,” https://www.apache.org/licenses/LICENSE-2.0; Mozilla, “Mozilla Public License Version 2.0,” https://www.mozilla.org/en-US/MPL/2.0/. ↩
Creative Commons, “Sharing Openly, Sharing Globally” (about the CC licenses, established in 2002), https://creativecommons.org/share-your-work/cclicenses/. ↩
Linux Foundation, “About the Linux Foundation,” https://www.linuxfoundation.org/about; Apache Software Foundation, “About the ASF,” https://www.apache.org/foundation/; Eclipse Foundation, “About the Eclipse Foundation,” https://www.eclipse.org/org/. ↩
Open Source Initiative, “History of the Open Source Initiative,” https://opensource.org/about/history-of-the-open-source-initiative (OSI founded in late February 1998); Open Source Initiative, “The Open Source Definition,” version 1.9, https://opensource.org/osd. ↩
Free Software Foundation, “FSF celebrates thirty-fifth anniversary with week of surprises and online event,” 6 October 2020, https://www.fsf.org/news/fsf-celebrates-thirty-fifth-anniversary-with-week-of-surprises-and-online-event (incorporated 4 October 1985); Free Software Foundation, “FSF History,” https://www.fsf.org/history (an initial free software definition appeared in the first GNU’s Bulletin in 1986). ↩
Free Software Foundation, “What is Free Software?,” GNU Project, https://www.gnu.org/philosophy/free-sw.html. ↩
From But Is It Open? by Timothy O’Brien (Discursive, 2026). © 2026 Timothy O’Brien. All rights reserved.