A Habr Essay Reopens the Debate Over Who Owns Open Source Code in the Age of AI

An essay titled "Open Source Is the Highest Form of Human Activity" has been published on Habr. The author opens with a confession: he has always been afraid of open source, because he has watched companies take other people's open work into closed commercial products, ignore licenses and snicker at the authors' naivety. In his view, this cannot end without consequences. The available excerpt contains no figures or specific cases, so only the thesis can be assessed. But it fits the last three years well: licensing disputes have moved out of legal footnotes and become part of company strategy.

Where the facts support the thesis

The first example is HashiCorp. In August 2023, the company moved Terraform from the MPL to the Business Source License, which restricts commercial use by competitors. The response came quickly: the community forked the project as OpenTofu and placed it under the Linux Foundation.

The second example is Redis. In March 2024, the project switched to dual licensing under RSALv2/SSPL, and Valkey appeared, backed by AWS, Google Cloud and Oracle. In 2025, Redis added the AGPLv3 license for version 8, a partial return to an open model. The market showed that closing a popular project is easier on paper than in practice: the community and large users can quickly assemble an alternative.

AI sharpens the question

In machine learning, the boundary is blurrier. Model code, weights and training data are legally distinct objects, and their licenses often do not match. The Doe v. GitHub lawsuit, filed in November 2022, challenged the training of Copilot on public repositories without complying with license terms. The court dismissed most of the claims, but the claims over license violations and contractual obligations remain in the case. There is still no final answer to whether a model can be trained on copyleft code without consequences.

On the other side are "open" models with strings attached. The Llama license requires a separate permission from Meta for companies with more than 700 million monthly active users, and it adds an acceptable use policy. The Open Source Initiative ultimately stated that such a license does not meet the definition of open-source software.

Key milestones:

• August 2023 — Terraform moves to BSL, and OpenTofu appears;
• March 2024 — Redis changes its license, and Valkey is born;
• 700 million MAU — the threshold in the Llama license beyond which a separate agreement is required;
• 28 October 2024 — the OSI publishes the Open Source AI Definition 1.0, under which openness requires access to code, parameters and a description of the training data;
• November 2022 — the Doe v. GitHub case begins over code in training datasets.

What teams and authors should do

Developers who embed open models and libraries in commercial products should do three things. First, keep a registry of dependencies with their licenses, including model weights and datasets, not just packages. Second, read the usage restrictions: audience thresholds, bans on specific use cases, attribution requirements. Third, for critical components, know in advance whether a viable fork exists under a neutral organization.

For authors of open projects, the takeaway is the reverse. If you do not want your code to quietly end up in closed products, choose a license deliberately: AGPL closes the network loophole, and a clear contribution policy lowers the risk of disputes. The Habr essay is emotional, but its point is practical: open code is not protected by moral right. It is protected only by the text of the license and the community's willingness to defend that license.

Note that this is an opinion column, not a release announcement. Legal conclusions about specific licenses are best checked with a lawyer.