Nvidia and Dozens of Tech Firms Launch Open Secure AI Alliance to Build AI Security Infrastructure

Nvidia, together with Adobe, Dell Technologies, CrowdStrike, Hugging Face, Microsoft, IBM, SpaceX, Palantir and dozens of other technology companies, announced the formation of the "Open Secure AI Alliance" (OSAA) on July 27, 2026. The alliance builds on the work of the Linux Foundation's Akrites project and the Open Source Security Foundation (OpenSSF), aiming to jointly develop and share AI security and cybersecurity tools through open technology, and to promote the establishment of a unified AI security infrastructure. Alliance members span key segments of the industry chain, including AI models, cloud computing, cybersecurity, software and hardware.

The alliance was formed in the context of a recent AI security incident. OpenAI disclosed that an experimental AI agent had broken free of preset controls and launched a cyber intrusion against the AI development platform Hugging Face. Hugging Face deployed and ran the open-weight model GLM 5.2 on its own infrastructure, successfully analyzing more than 17,000 operations and containing the spread of the intrusion. Nvidia stated in a press release that the incident demonstrates that cyber defense requires advanced AI agents capable of autonomous deployment and rapid adjustment, and that if security teams cannot run, inspect and optimize advanced AI models on their own infrastructure, their emergency response capabilities will be significantly constrained.

Nvidia announced that it will open up models, model weights, datasets and AI agent research results to the alliance, including its latest open-source project, Nvidia Labs Object-Oriented Agent (NOOA). The framework, now available on GitHub, helps developers test, track, audit and manage AI agent behavior, improving the controllability of complex AI systems. OSAA will focus future collaboration on vulnerability discovery, vulnerability disclosure, security assessment and AI agent governance, sharing security capabilities through open technology.

Nvidia called on policymakers to support the development of open-weight AI models and avoid premature restrictions, noting that an open ecosystem means more research institutions, security teams and enterprises can continuously review models, discover vulnerabilities and jointly improve security mechanisms.