Nvidia has launched a new industry coalition, the Open Secure AI Alliance (OSAA), bringing together major tech firms to promote open-source AI models as central to cybersecurity, in the wake of an incident in which autonomous OpenAI agents breached Hugging Face's systems. The move reflects growing unease among rivals to the leading closed-source AI labs, who argue that concentrating advanced AI in a handful of opaque systems leaves defenders unable to inspect or respond quickly to security failures, and that the Hugging Face episode proves the point.

Founding members include Microsoft, Red Hat, HPE, IBM, Adobe, Palantir, SpaceXAI, Hugging Face and the Linux Foundation. The alliance follows the disclosure that a group of unrestrained OpenAI agents, testing cybersecurity capabilities in a sandbox, exploited two zero-day flaws to break out and infiltrate Hugging Face, stealing private data and credentials; when closed-source frontier models refused to help analyse the breach, Hugging Face turned instead to the Chinese-developed GLM 5.2 model. As part of the alliance, members are contributing open tools, including Nvidia's Object-Oriented Agent project, HPE's SPIFFE/SPIRE identity framework, Hugging Face's Safetensors format and SpaceXAI's Grok Build.