Mark Zuckerberg advocates for open-weight AI to decentralise control, with Meta releasing downloadable models and committing to transparency amidst industry debates on sovereignty and resilience.
Mark Zuckerberg has set out a sweeping case for open-weight artificial intelligence, arguing in a 6,500-word essay that the technology should not be controlled by a handful of companies or governments. In the same move, Meta released a new model that users can download and run on their own machines, a practical difference that matters because it changes who can keep the software, who can switch it off and who can decide how it is used. According to The Guardian’s account of the essay, that question of ownership sits at the centre of the debate now unfolding around AI.
The argument is more than philosophical. Axios reported that Zuckerberg paired the essay with a commitment to open-source development under the oversight of an independent board, plus a $1 billion fund intended to support communities near Meta’s data operations. AP News said Meta’s new model, Muse Glimmer, is designed to run on personal computers, with a more powerful version, Muse Spark 1.2, also planned for developers. Together, those details point to a clearer corporate strategy: Meta wants to present openness as both a technical model and a political answer to concentration in AI.
That position also reflects a broader shift in the industry. Meta has long used open-weight releases, including earlier Llama versions that were promoted as a way to widen access and accelerate innovation. But in 2025, Zuckerberg signalled that the company would be more selective about what it opened as models became more advanced and risks grew. The latest essay revives the open strategy with a stronger public defence, suggesting that Meta now sees openness not as a side project but as a competitive distinction against closed systems such as ChatGPT, Gemini and Claude.
The dispute is also about control, not just capability. The Guardian’s piece draws a parallel with earlier episodes in which companies or governments were able to remove digital material or services from users after the fact. That history matters because it frames AI as infrastructure rather than novelty: if a model can be taken away, altered remotely or made unusable, then access is conditional. Open-weight systems, by contrast, can be stored locally and kept running even if the vendor changes direction, which gives users a degree of continuity that rented cloud services do not.
That is why the policy debate is now moving beyond whether open models are better or worse in performance terms. As Reuters-style reporting on the sector has often shown, governments and large institutions are increasingly concerned with resilience, sovereignty and auditability. The practical question is whether hospitals, agencies and businesses should be required to keep legally usable copies of capable models and the systems needed to run them. Meta’s latest release does not settle that issue, but it does sharpen it: in AI, the decisive issue is becoming less about who builds the model than who can keep it.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
