Nvidia launches Open Secure AI Alliance with 50+ members. OpenAI, Google, and Anthropic sit out.

Nvidia launched a security coalition of more than 50 companies to build open tools for AI agent defense. OpenAI, Google, and Anthropic are not members.

"Attackers have frontier AI. Defenders need a frontier AI ecosystem," Nvidia CEO Jensen Huang said on X. The launch directly referenced the recent Hugging Face security incident, where closed AI models blocked forensic analysis of an attack because the logs contained exploit payloads.

The alliance is not a standards body. It is a working coalition shipping open tools. Nvidia contributed NOOA, a research framework that lets agent harnesses trace, audit, and govern AI model behavior. Microsoft contributed MDASH, a multi-model agentic scanning system for vulnerability discovery. Hugging Face offered Safetensors, a model format that prevents hidden code execution.

The absence of the three largest frontier labs is the story within the story. Nvidia has not said whether OpenAI, Google, or Anthropic were invited. All three have investments in closed-model security infrastructure. Their business models depend on API access, not open tools.

The alliance frames open-weight models as defensive assets, not security liabilities. "When defenders cannot inspect, adapt and run advanced AI on their own infrastructure, their ability to respond is constrained at exactly the moment speed matters most," the announcement said.

The moment is timely. Huang's defense of Chinese open-source AI models, the Hugging Face incident, and the alliance launch all land within the same week. The pattern is clear. Nvidia is building an open defense ecosystem that runs on Nvidia hardware. The closed-model labs are not invited. Their customers are.

One loose thread: the alliance has no announced governance structure yet. Fifty-plus companies with competing interests and no chair, no working groups, no roadmap. The tools are real. The coalition's staying power is unproven.

Source