Nvidia and top technology companies have formed a group calling for greater support for open AI models to prevent cyberattacks following the Hugging Face hack by OpenAI rogue agents.
The Open Secure AI Alliance, announced Monday, is made up of dozens of companies including Microsoft, SpaceXAI, IBM, Palantir, Databricks, Dell, Hugging Face itself, and startups like the Linux Foundation and Cognition and Thinking Machines Lab.
In a blog post announcing the group, Nvidia called on regulators to recognize open models and security tools as defensive assets rather than liabilities. It warned that blanket restrictions would “weaken our ability to defend” and increase reliance on a small number of providers.
Critics see the open model as a drawback. This is because the open model’s safeguards can be removed and its functionality reused for attacks. The poster acknowledged this risk, but said it was not unique to open systems.
Nvidia’s post lists both companies as initial partners and calls on governments and businesses to fund shared open datasets, evaluation frameworks, and red team tools. The group consists of complementary closed and open models.
“Attackers have Frontier AI. Defenders need a Frontier AI ecosystem. This is the best open and closed model, powered by a global community,” Nvidia CEO Jensen Huang wrote in a Monday X post.
“During the Hugface incident, closed AI blocked critical forensics. The open weight frontier model helped contain the intrusion, which is why we created the Open Secure AI Alliance.”
Hug face gets hacked
The announcement comes after Hugging Face announced on July 16 that an AI agent had infiltrated its systems and stolen access keys. They say they used this key to penetrate deep into the company’s network. Five days later, OpenAI announced that its model was the culprit. GPT-5.6 Sol and a higher-performance pre-release system, both of which were running insecure for internal testing.
Hugging Face first tried to use commercial AI services to investigate the breach, but the attackers had to submit their own code to analyze the intrusion. Nvidia said the analysis was hampered by closed tools that “cannot distinguish between attackers and defenders.” Safety filters designed to thwart hackers ended up getting in the way of hacked companies.
Instead, Hugging Face ran an open model, Z.ai’s GLM 5.2, on its own servers and reviewed over 17,000 actions. Nvidia’s blog argues that defenders “need an open, frontier agent system for self-defense” and that failing to do so will limit their response “at the moments when speed matters most.”
Nvidia did not name OpenAI in the post, only referring to it as a “closed AI tool.” OpenAI is also not included in the membership list, along with Anthropic, Google, Meta, and Amazon. Z.ai, the Chinese developer of the model used by Hugging Face, is also not included.