NVIDIA announced on July 27, 2026 that it is joining a new group called the Open Secure AI Alliance. The alliance plans to build and share open source tools for AI safety and security.
More than 50 organizations signed on as founding members. The list includes tech and cybersecurity companies such as Microsoft, IBM, Cisco, Adobe, Salesforce and Palo Alto Networks.
The alliance builds on earlier work from the Linux Foundation's Akrites initiative and the OpenSSF community. Its goal is to find and fix security flaws using open source methods.
The group says open source software already plays a large role in cloud computing, banking, manufacturing and government systems. It argues that AI security should follow a similar open model.
Why the Alliance Was Formed
The announcement points to a recent security incident at Hugging Face as one reason for the new effort. During that incident, closed AI tools could not tell the difference between attackers and defenders.
This blocked forensic analysis that Hugging Face needed to investigate the breach. The company instead used an open weight model called GLM 5.2 on its own servers.
That model reviewed more than 17,000 actions and helped contain the intrusion. The alliance says this shows why open tools matter for defenders.
The group argues that when companies cannot inspect or run AI systems on their own infrastructure, their ability to respond to attacks is limited. It says this is especially true during fast moving incidents.
What the Members Are Contributing
NVIDIA is giving the alliance open models, model weights, data and new research on agent harnesses. A harness is the system that connects an AI model to the tools and rules it needs to work safely.
NVIDIA also released a new open source project called NOOA, short for NVIDIA Labs Object Oriented Agent. It is now available on GitHub and is meant to make AI agents easier to test, trace and monitor.
Other members are contributing their own tools. HPE is working on SPIFFE and SPIRE, a system that verifies the identity of AI agents before they can access company data.
Hugging Face has given a tool called Safetensors to the PyTorch Foundation. It stores AI model weights in a format designed to prevent unwanted code from running.
IBM and Red Hat's Lightwell project adds digitally signed patches to the open source supply chain. Microsoft contributed a scanning tool called MDASH that uses multiple AI agents to find and confirm software bugs.
SpaceXAI has open sourced its Grok Build coding agent. The company also plans to release the weights for its Grok line of models to researchers and developers.
The alliance is also asking policymakers to treat open AI systems as tools for defense rather than risks to be restricted. It says blanket rules against open frontier models could weaken cybersecurity efforts overall.
The group wants governments and companies to invest in shared resources such as datasets, evaluation tools and attack simulators. These would be similar to public investments made in open source software in the past.
As of the July 27 announcement, the Open Secure AI Alliance is inviting more governments, companies and researchers to join. NVIDIA said the NOOA project is already live on GitHub for public use.
