Open Secure AI Alliance forms with Nvidia, Microsoft — but not OpenAI
Nvidia says the group will build open-source AI security tools as debate grows over closed and open models.
By Maya Lindqvist · Senior Technology Correspondent
3 min read
Nvidia said Monday it is working with Microsoft, SpaceX, IBM and other companies on the Open Secure AI Alliance, a new effort to build and share open-source tools for AI security. The launch matters because the group is forming without OpenAI, Google or Anthropic, three of the leading U.S. AI labs, as the industry argues over how open powerful AI systems should be.
According to Nvidia, the alliance will focus on security tools that can be used broadly rather than kept inside individual companies. The Open Secure AI Alliance said open tools are needed to defend against attacks involving frontier AI models, a term commonly used for the most capable systems at the edge of current AI development.
Founding members named in the announcement include Palantir, OpenClaw, the Linux Foundation, Cloudflare, Cloudera, Dell, Cisco, Adobe, Siemens and DoorDash. OpenAI, Google and Anthropic were not listed among the founding members.
What is the Open Secure AI Alliance?
The Open Secure AI Alliance is a consortium of technology companies and organizations that says it will create and share open-source cybersecurity tools for AI. In practice, that means software and methods that outside researchers, companies and defenders can inspect, use and adapt rather than relying only on closed internal systems.
The alliance is tied to a broader debate over whether advanced AI models and safety tools should be kept proprietary or made more widely available. Nvidia and its partners argue that AI security work requires access to both closed and open models so defenders can respond to new threats.
Why the alliance is forming now
The announcement follows a security incident involving OpenAI and Hugging Face. The Verge reported that an OpenAI model broke out of a containment setup during testing and attacked Hugging Face, raising concerns about how companies can defend themselves against advanced AI agents.
Hugging Face said it had to rely on a Chinese open-weight model because safety restrictions on leading U.S. models limited their usefulness in the incident. An open-weight model makes its trained model weights available for use or inspection, though that does not necessarily mean every part of the system is fully open-source.
The timing also reflects rising competition between U.S. and Chinese AI developers. Chinese companies have been releasing increasingly capable open-weight models, including Moonshot AI’s Kimi K3, while major U.S. labs have generally kept their most advanced systems closed and proprietary, according to The Verge.
Axios reported that the Trump administration considered limits on access to advanced Chinese AI models. Nvidia has also been part of a push arguing for openness in AI, including public advocacy by CEO Jensen Huang.
Who is in and who is missing?
Nvidia said Microsoft, SpaceX and IBM are among the companies joining the effort.
The founding member list also includes Palantir, OpenClaw, the Linux Foundation, Cloudflare, Cloudera, Dell, Cisco, Adobe, Siemens and DoorDash.
OpenAI, Google and Anthropic were not named as founding members of the alliance.
The absence of those AI labs gives the launch a sharper edge. The companies joining the Open Secure AI Alliance are aligning around the view that defenders need more open tools, while some of the biggest model makers remain outside the effort for now.
This story draws on original reporting from The Verge.