After an AI model went rogue, Nvidia decided it was time to get serious about safety. The chip giant is pulling together 37 companies — including Microsoft, SpaceX AI, Hugging Face, and IBM — to form the Open Secure AI Alliance. The group's mission: develop open-source models, agent runtime frameworks, and security tools that can spot vulnerabilities, audit agent behavior, and defend against AI-powered attacks. Notably absent from the initial roster: OpenAI, Google, and Anthropic.
Jensen Huang, Nvidia's CEO, pointed to the recent breach at Hugging Face as the direct catalyst. "Attackers already have frontier AI," he said. "Defenders need a frontier AI system made up of the best open and closed models, powered by the global community."
Alongside the alliance, Nvidia is open-sourcing NOOA, an agent framework that lets you write agents as single Python classes — making them easier to test, trace, audit, and control.