Nvidia launches AI initiative as OpenAI cyber attack fallout continues
Nvidia As the fallout from a cyberattack carried out by fake OpenAI models continues, a number of tech giants on Monday launched a new AI security initiative focused on open models.
Last week, it was revealed that the target of the attack, the startup Hugging Face, was unable to use leading US border models to defend itself, and that the guardrails did not distinguish between attacker and defender. Instead, it has turned to a self-hosted, vulnerability-heavy Chinese model that is not bound by the same constraints.
Tech giants have launched an initiative aimed at creating and sharing open AI tools after US lawmakers increasingly weighed how to stem the growing adoption of Chinese AI models, the most advanced of which are open heavyweights.
Unlike closed models, including edge systems created by Anthropic and OpenAI, that can only be accessed through a specific infrastructure, open models can be downloaded, modified, and self-hosted.
“The Open Secure AI Alliance will work to fix and expose vulnerabilities using open technologies,” Nvidia said in a statement. “The recent Hugging Face security incident provides a stark reminder: Cyber defenders need open, edge-operating agent systems for self-defense.”
Besides Nvidia, other members of the alliance include: Microsoft, SpaceX, palantirand dozens of other tech companies from the US and Europe.
Effort to block Chinese artificial intelligence
There are growing calls for measures to limit access to models created by Chinese artificial intelligence companies. Campaigns to extract information from rivals’ systems in the United States, known as “distillation,” meaning a model extracts information from a better-trained model.
Last week, Treasury Secretary Scott Bessent threatened sanctions against Chinese companies that conduct distillation attacks on U.S. companies.
“There is a real possibility that the U.S. government will clamp down on Chinese models,” Chris McGuire, senior fellow for China and emerging technologies at the think tank Council on Foreign Relations, told CNBC.
This could include banning transactions that involve models such as purchasing tokens through an API or charging customers for inference by US companies that host the model in the cloud, he said.
“In Washington, this is not an open source versus closed source debate, it’s a debate about whether intellectual property theft in China should be tolerated,” McGuire said. “Any action will focus on Chinese companies, not the open source ecosystem.”
But there are concerns about restrictions because the majority of the most capable open source models are produced by Chinese companies.
Last week Nvidia, Microsoft, MetaPalantir and more than 20 other companies published a letter urging policymakers to avoid “premature restrictions” on vulnerability-heavy AI models that would “stifle competition or encourage innovation abroad.”
Nvidia said the OpenAI-Hugging Face incident demonstrated a “practical truth.” “When defenders cannot control, adapt and run advanced AI within their own infrastructure, their ability to respond when speed matters most is constrained.”
