Open Source and AI
The AI hype cycle may be fading, but its impact on Open Source is just beginning. AI-generated contributions are flooding repositories, shifting responsibility onto maintainers who must now validate, debug, and assess licensing risks.
Beyond contribution quality concerns, Open Source communities are debating whether AI-generated outputs should be considered derivative works of the Open Source code they were trained on. If an AI model was trained on GPL, should its output require the same license? Does it even count as a derivative work? Some projects are outright rejecting AI-generated pull requests out of legal uncertainty. A lot of it is to be decided in legal courts, and countries have different stances with respect to copyright implication on a model training.
This debate extends to what “Open Source AI” should mean. The Open Source Initiative (OSI) recently introduced its Open Source AI Definition (OSAID) after a long community feedback process. OSI has allowed for some leeway with training data in the name of practicality. They see this as a practical compromise to encourage Open AI development, but a big part of community argue it undermines reproducibility: a core Open Source principle.
The Digital Public Goods (DPG) AI Community of Practice (CoP) recommendation takes an even stricter stance, recommending that AI solutions cannot qualify as Digital Public Goods unless their training data is fully open. This reflects a deeper divide: should Open Source AI allow partial transparency, or must all components including data be freely available to ensure accountability?
The Software Freedom Conservancy (SFC) has gone further, criticizing OSAID as a corporate-driven compromise that lets AI vendors retroactively label proprietary systems as "Open Source." SFC argues that Open Source AI must guarantee full reproducibility not just disclose methodology but provide all necessary components (including training data) to rebuild models independently. Additionally, AI raises broader ethical concerns, as many generative models train on creative works without consent from artists, writers, and musicians.
This growing divide raises critical questions: Will AI governance frameworks converge, or will multiple competing definitions emerge? Is Open Source evolving to accommodate AI, or is it being reshaped by corporate influence? Just as the Open Source Definition (OSD) has evolved over two decades, Open Source AI will likely need continuous refinement to balance openness, ethics, and sustainability.
We also notice the ever growing “open washing” specifically prominent with LLMs. Models like Meta’s Llama and DeepSeek R1 are pretending to be Open Source but they don’t meet even the conservative definition by OSI.