Zuckerberg calls for change: Open-source artificial intelligence key to fight China

The new Muse Glimmer model is significantly smaller than the leading AI models of rival companies and is intended for agent artificial intelligence, and can be run on personal computers with only one graphics card.

|
Video: 24sata/reuters

Open source models are generally cheaper than leading models from companies like OpenAI and Anthropic. Also, it is easy to adapt them to personal needs, unlike closed models that companies keep completely under their control.

The announcement of the new Meta model comes at a time when the company is trying to consolidate its position after last year it established an expensive superintelligence research team to get back into the race with other companies developing new technology.

Zuckerberg’s comments also represent the latest support for open-source artificial intelligence, which is gaining momentum as companies wary of the rising costs of AI, but also recent cybersecurity incidents involving Anthropic, OpenAI and Meta models.

American company Hugging Face, which was hacked by OpenAI’s model, said last month that it used an open-source Chinese model to defend against attacks because closed-source models have cybersecurity limitations.

The value of Meta’s shares, which have fallen by around 10 percent this year, rose by one percent on Monday before the market opened.

Necessary revision of policies

In a press release, Zuckerberg said that the United States must reconsider its policies if it wants domestic companies to lead when it comes to about open source models.

Chinese development companies lead the race of open source models, and models such as Moonshot’s Kimi K3, Alibaba’s Qwen3.8-Max, and DeepSeek’s V4-Flash compete with the most powerful systems of American companies. The leading models of the American companies OpenAI, Anthropic and Alphabet are closed source.

“US policies must reduce these additional barriers if we want American open source models to lead the way over time,” Zuckerberg said, adding that restricting access to foreign models is not an effective solution.

US President Donald Trump’s administration told AI development engineers earlier this month that it would not require security testing of open-source models, according to two people familiar with the matter.

In the press release, Zuckerberg also advocated the so-called distillation of AI, that is, the use of more powerful artificial intelligence systems to train a smaller model.

He said that Meta will introduce a governance structure in which independent directors will have the authority to approve the security criteria for model publication.