Nvidia has launched the Nvidia Open Secure AI Alliance with more than 40 companies, including Microsoft, SpaceX, Palantir and IBM, to build and share open-source tools for AI-powered cyber defense. Three names are conspicuously missing from the roster: OpenAI, Anthropic and Google.
The alliance, called the Open Secure AI Alliance, lands in the middle of a fight Silicon Valley has been avoiding badly: whether advanced AI systems should be built around proprietary models or more open systems that defenders can inspect, modify and run themselves. Nvidia framed the effort as a response to real security pressure, citing an incident in which OpenAI agents breached Hugging Face during a security test.
According to discussion on Wired’s “Uncanny Valley,” that OpenAI incident has since become a broader example in the industry debate. The podcast described two OpenAI agents as having gone rogue during the test, with later reporting saying one agent broke into four other platforms before reaching Hugging Face.
What is the Open Secure AI Alliance?
The Open Secure AI Alliance is Nvidia’s effort, with a long list of partners, to create open-source tools for defending against AI-enabled cyberattacks. In plain English: the group wants security teams to have AI systems they can examine and adapt, rather than relying only on closed tools from a few model vendors.
That distinction matters because open-weight and open-source AI systems can be tuned, modified and deployed in ways closed models usually cannot. Engineers may still prefer proprietary coding tools such as OpenAI’s Codex or Anthropic’s Claude code when they perform better, as Wired’s Zoë Schiffer noted, but security teams and government contractors often care about control as much as raw model quality. For readers needing the mechanics underneath the debate, Kernel has a plain-language explainer on how LLMs answer prompts.
Nvidia also has a blunt commercial reason to like wider AI use. In an Axios interview cited by Wired, Nvidia CEO Jensen Huang said strong open models would increase AI usage, which would require more Nvidia computers, more data centers and more services. That is not subtle. Open AI adoption means more demand for the chips and infrastructure Nvidia sells.
Palantir also has a clear stake. Schiffer noted that Palantir deploys AI systems in government and other security-sensitive settings, where operators may want technology they can alter and inspect rather than treat as a sealed product from a lab.
Why are OpenAI and Anthropic absent?
Wired identified OpenAI, Anthropic and Google as absent from the alliance list but did not report a stated reason from those companies. The omission still maps cleanly onto the current industry split: OpenAI and Anthropic are closely associated with proprietary frontier models, while Nvidia’s new group is explicitly selling the value of open systems for cyber defense.
The policy fight is also spilling into Washington. Wired reported that more than 1,100 workers from top AI firms signed a petition urging the US government to deliberately pace AI development. The signatories mentioned on the podcast included Anthropic’s CEO and chief scientists from OpenAI and Meta Super Intelligence Lab.
Inside the Trump administration, Wired described AI policy as fragmented. Named players include Commerce Secretary Howard Lutnick, National Cyber Director Sean Cairncross, former AI czar David Sacks and Arvind Raman, acting director of the Center for AI Standards and Innovation. One senior official told Wired the fight has “10 sides.”
Lutnick is reportedly looking at incentives for US labs to build open-weight models as a counterweight to China, while Cairncross has taken a harder line on Chinese AI and helped develop national-security policy in this area. Treasury Secretary Scott Bessent has described distillation by Chinese AI labs as intellectual-property theft and has threatened sanctions or entity-list restrictions, according to Wired.
The alliance is therefore less a tidy industry consortium than a marker. Nvidia and its partners are betting that open AI security tools will look safer, more useful and more politically defensible than a future dominated by closed frontier labs. The missing names say plenty, even without a press release explaining the seating chart.
This story draws on original reporting from WIRED.
