NVIDIA announced the establishment of the Open Secure AI Alliance, partnering with founding members including Adobe, Microsoft, Hugging Face, IBM, Linux Foundation, Cloudflare, CrowdStrike, Hewlett Packard Enterprise (HPE), Palantir, Salesforce, SAP, and SK Telecom to jointly develop and share open technologies for AI security and cyber defense. The alliance asserts that AI safety should extend beyond whether model weights are open, encompassing an entire technical stack that includes agent identity, permissions, isolation mechanisms, guardrails, logging, and evaluation. By making models and frameworks openly accessible, defenders can inspect, modify, and deploy them locally, thereby reducing reliance on a small number of proprietary vendors. At the same time, risks associated with using these open systems for network attacks or removing safety restrictions must be managed through rigorous assessment and usage protocols. NVIDIA will contribute open models, weights, data, and agent framework research to the alliance and launch NOOA, an open-source project designed to help developers test, track, audit, and govern agent behavior. Other member organizations will advance initiatives in zero-trust authentication, secure model formats, software supply chain protection, and multi-model vulnerability scanning. The alliance also calls upon regulators to treat open AI as a defensive resource rather than imposing one-size-fits-all restrictions, while investing in shared datasets, attack simulators, evaluation frameworks, and red-teaming tools.