NVIDIA and dozens of industry partners — including Microsoft, IBM, Red Hat, Hugging Face, CrowdStrike, and Cloudflare — have launched the Open Secure AI Alliance, aimed at building and sharing open tools, models, and agent harnesses for AI-era cybersecurity defense. The initiative argues that open frontier AI models are essential for defenders, citing a real Hugging Face security incident where open-weight models enabled forensic analysis that closed tools blocked. Contributions include NVIDIA's NOOA agent harness framework, Hugging Face's Safetensors format donated to the PyTorch Foundation, IBM/Red Hat's Lightwell supply chain security, Microsoft's MDASH multi-model scanning harness, and SpaceXAI open-sourcing Grok Build. The alliance also calls on policymakers to treat open AI models as defensive assets rather than liabilities.