AI open‑source vs open‑weight definitions spark industry and geopolitical debate

The terms “open source” and “open weight” are often conflated in the artificial‑intelligence sector, leading to what experts call “open‑washing”. In 2024 the Open Source Initiative released a stable definition (OSAID) that requires not only the model’s weights but also the training code and data, or a full description of its composition, to qualify as truly open source.

Industry players such as Meta have faced criticism for labeling models like Llama 3 as open source without meeting these criteria, prompting a shift toward the broader label “open model”. The distinction matters for sovereignty and security: the United States briefly blocked access to certain models for national‑security reasons, while China is contemplating restrictions on its own models to prevent foreign exploitation. Clarifying the definitions is seen as essential for transparency, independent audit, and preventing concentration of AI power.

The debate underscores growing geopolitical stakes around AI governance and the need for clearer standards to guide developers, regulators, and the public.