{
  "id": 7511872,
  "title": "Open weights are not open source: Why AI's favorite label is under dispute",
  "url": "https://urgent.news/2026/09/15/open-weights-are-not-open-source-why-ais-favorite-label-is-under-7511872",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-15T08:30:00.000Z",
  "source": {
    "name": "The Register",
    "slug": "the-register",
    "url": "https://www.theregister.com/columnists/2026/09/15/open-weights-are-not-open-source-why-ais-favorite-label-is-under-dispute/5295436"
  },
  "original_language": "en",
  "account": "The AI industry often misuses the term \"open.\" Companies publish model files on platforms like Hugging Face, and developers run them on their own GPUs. They label this as an \"open source model,\" but this may not be entirely accurate. Open weights, which are the learned numerical parameters from training, are publicly available. You can download them, self-host the model, fine-tune it on internal documents, and avoid using a proprietary API. Open weights offer greater control over data, privacy, costs, and vendor lock-in. However, releasing weights alone does not provide a complete picture of the model, as it does not disclose the training data or documentation. This lack of transparency prevents others from fully inspecting, reproducing, altering, and redistributing the system. The Open Source Initiative (OSI) recognizes the difference between open weights and open source AI. Open weights merely allow you to run a model on your own machine, while open source AI enables you to trust the model, enhance it, and build upon it. Without disclosing training data or providing detailed documentation, outsiders cannot fully test, reproduce, or challenge the model. The OSI has its own definition of open source AI, which requires model parameters to be made available under OSI-approved terms, but does not prescribe a specific legal mechanism. However, some prominent figures in the open source community have criticized the OSI's Open Source AI Definition (OSAID) 1.0, arguing that it leaves a significant gap in fully open AI systems. Bruce Perens, the author of the original Open Source Definition, denounced OSAID as not being open source. Others, like Luca Antiga and Bradley Kuhn, have called for the repeal of OSAID, citing its negative impact on the free and open-source software (FOSS) community and the OSI's reputation. Despite these criticisms, the Linux Foundation has submitted the Open Model, Data, and Weights (OpenMDW) license to the OSI. OpenMDW aims to define separate terms for a model's architecture, training data, and weights, addressing the limitations of current licensing approaches. However, the OpenMDW license has faced objections from some OSI members, including Stefano Maffulli, who suspects an ideological bias against big tech and AI. Landay emphasizes that while open weights enable running models locally, open source AI is essential for building trust, improving models, and fostering innovation. The current situation shows that almost everyone can run AI models, but few are able to trust, enhance, and build on them. There is a need for both open weights and open source AI to ensure a truly open AI ecosystem.",
  "summary": "Downloading a model is increasingly easy. Understanding how it was made, or changing a system at its root, is another matter",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 2,
    "also_reported_by": [
      {
        "outlet": "The Register Science",
        "title": "Open weights are not open source: Why AI's favorite label is under dispute",
        "url": "https://urgent.news/2026/09/15/open-weights-are-not-open-source-why-ais-favorite-label-is-under",
        "published": "2026-09-15T08:30:00.000Z"
      }
    ]
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}