NVIDIA and a coalition of industry partners launched the Open Secure AI Alliance on July 27, 2026, framing the future of AI security as a binary choice between opaque, proprietary defenses and open, community-driven tools. The group identifies cybersecurity as one of the top three fields that will benefit from this open approach, directly affecting IT and development teams responsible for protecting critical infrastructure.
Open models aim to democratize defensive capabilities
Building on the Linux Foundation's Akrites initiative and the OpenSSF community, the alliance plans to share open technologies that remediate vulnerabilities and build trust in AI systems. "The world needs both closed and open models," the alliance said, stressing that open models are essential for widening access to defensive tools and avoiding single points of failure.
The alliance emphasizes cybersecurity as one of the top three areas to gain from open AI defense, a focus that aligns with the growing need for security professionals to integrate AI into their workflows. This need is reflected in specialized training paths like the AI Learning Path for Cybersecurity Analysts.
Industry contributions and the NOOA project
Contributing members include NVIDIA, Adobe, and Microsoft, which are developing identity, permissions, and secure coding workflows. HPE is working on zero-trust identity frameworks through SPIFFE/SPIRE, and Hugging Face contributed Safetensors, a secure format for storing model weights, to the PyTorch Foundation.
NVIDIA is also releasing open models and research via the NOOA project, designed to make AI safety capabilities more accessible for agent harnesses. The framework, hosted on GitHub, aims to improve how harnesses integrate with models, making agent behavior easier to test, trace, audit, and govern. The alliance argues that combining openness with rapid remediation will prioritize shared security over secrecy.
Hugging Face breach exposes limits of closed systems
A recent security breach at Hugging Face underscored the value of open systems. When attackers gained access, Hugging Face used the open-weight GLM 5.2 model on its own infrastructure to analyze over 17,000 actions and contain the threat-a response that would have been impossible without direct access to the model's internals.
"The safer path is the one that gives more defenders the ability to test, verify and strengthen the systems on which society relies," the alliance said. The incident reinforces the argument that open, verifiable AI tools create a distributed defense system that can react faster than a handful of closed vendors.
Why this matters for IT and Development
For IT and development teams, the shift toward open, verifiable AI security means building systems that can be independently audited and adapted under pressure. Professionals who can work with open models, agent frameworks, and zero-trust identity protocols will be better positioned to protect their organizations. Resources like AI for IT & Development cover the skills needed to deploy and maintain these emerging defenses.
Your membership also unlocks:
