Big techs defend open AI after Chinese model interrupts attack caused by OpenAI error

The letter released this Friday (24) is signed by Nvidia, Microsoft and Meta and contradicts claims by OpenAI, Anthropic and Google of limits to Chinese technology, although there is no direct mention of the country in the document.

While open language models can become weapons in the hands of attackers, the same technology "expands defensive capabilities, increases transparency, and allows vulnerabilities to be discovered and remediated," the letter says.

The manifesto follows the attack reported by Hugging Face on the 16th. The site, which is the main library of AI models on the internet, only managed to expel the uncontrolled OpenAI robots that invaded its systems after resorting to the open source Chinese GLM 5.2 model. Previously, he tried, unsuccessfully, to use cutting-edge American technology.

"Analysis requires processing large volumes of attack commands, infrastructure vulnerabilities, and artifacts, and these requests were denied due to security safeguards at key labs," the company explained.

"The choice for the open model had a second advantage: none of the attack codes nor the credentials [passwords and users] mentioned left our systems", added Hugging Face in a report.

Hugging Face's report is in line with this Friday's letter. While concentrating advanced AI in a small group of closed models increases risks, "open models allow a broad group of researchers to examine technology behavior, identify vulnerabilities and develop appropriate safeguards over time", argue the big tech signatories.

In his first tweet, Nvidia CEO Jensen Huang states that "open models strengthen defense and cybersecurity, accelerate innovation and popularization, enabling sovereignty."

Huang's statement refers to the recent speech by Chinese leader Xi Jinping, who, on the 17th, defended joint work on technological development.

"The development of AI should not be the work of a single country, but rather a symphony of international cooperation," he said at the opening ceremony of the World Artificial Intelligence Conference (WAIC).

China has already prioritized open source models since Alibaba announced the first version of Qwen in 2023. The DeepSeek R1, the first Chinese model to shake up the market, in January 2025, is also open source.

The situation repeats itself this month, with the announcement of Kimi K3, from the Moonshot laboratory, positioned among the three most powerful AIs on the market, alongside technologies from the American OpenAI and Anthropic.

The Kimi K3 became an example around the world of the strength of open models and a victory for Chinese engineering, despite the restriction on exports of American chips to the Xi Jinping regime. The model is optimized to run on Chinese hardware in inference, such as Huawei chips.

Moonshot's report shows that the model was partially trained with Chinese chips, indicating that foreign equipment was also used. The text is more detailed than any descriptive document on artificial intelligence from the main American laboratories.

Moonshot should make the editable version of the model and its weights available next Monday (27), when companies and people, with access to supercomputers, will be able to re-edit their own versions of the AI, with different specialties.

The claim by American companies against the restriction of Chinese technology and in favor of open technology has geopolitical and economic objectives, according to the advisor of CGI.br (Internet Steering Committee in Brazil) Rodolfo Avelino.

"If the US excessively restricts open models, it could reduce its capacity for distributed innovation, as today innovation is concentrated more in large corporations like Google and Anthropic than in universities," he said.

For Avelino, American companies also have economic interests in proposing a preference for open models. Meta and Nvidia provide their own open models and Microsoft profits when companies use its open infrastructure to run heavy AI models, whether Chinese or American.

"This letter expresses a dispute over who will capture value in the next phase of the artificial intelligence economy", says the CGI.br advisor.

"We can have an open source model and continue to depend on the companies that control the chips, the data centers and the cloud. That's where the power remains concentrated."

SEE WHO SIGNED

1. American Innovation Network2. Andreessen and Horowitz3. Aceee4. Arena5. Black Forest Labs6. Box7. Crowdstrike8. Dell9. Emergence10. Hugging Face11. IBM12. The Linux Foundation13. Mariana Minerals14. Goal15. Microsoft16. Mistral AI17. Mozilla Foundation18. Nvidia19. Palantir20. Perplexity21. Reflection22. Replit23. ServiceNow24. Telnyx25. Y Combinator

Read Also: EU points out possible violation of digital legislation by TikTok in the protection of minors