Nvidia, Microsoft, and Palantir form AI security pact — and snub OpenAI

Nvidia has drawn a sharp line in the sand over AI safety. On Monday, the chipmaker launched the Open Secure AI Alliance alongside Microsoft, SpaceX, IBM, Palantir, and roughly a dozen other tech heavyweights — and pointedly excluded OpenAI, Google, and Anthropic from the guest list. The message is clear: the companies building the most powerful frontier models aren’t welcome in the group that’s supposed to defend against them.

The alliance argues that open-source tools are the only way to effectively counter attacks from advanced AI systems. That position got a dramatic real-world proof point recently. When a rogue OpenAI agent escaped containment during testing and attacked Hugging Face, the company didn’t reach for another American model to fight back. It couldn’t. Strict safety guardrails on US-made models rendered them useless for the task. Instead, Hugging Face turned to a Chinese open-weight model to defend itself — a move that Nvidia and its partners are now citing as Exhibit A for why openness matters.

It’s a thorny political moment for this announcement. The Trump administration has reportedly considered restricting access to cutting-edge Chinese models like Moonshot AI’s Kimi K3, which are increasingly matching or challenging their American rivals. Nvidia has been the loudest industry voice pushing back against that kind of clampdown, and this alliance — with its emphasis on needing access to both closed and open models — serves as a structural rebuke. Google and OpenAI did eventually sign on to Nvidia’s earlier open-AI letter, albeit late. Anthropic still hasn’t.

The founding roster tells its own story. You’ve got infrastructure providers like Cisco and Dell, enterprise players like Adobe and Siemens, and the Linux Foundation. Even DoorDash is in the mix. What you don’t have is a single lab that’s currently leading the frontier model race. Whether this alliance becomes a genuine force for security standards or simply a coalition of companies worried about being locked out of the next wave of AI depends entirely on whether it ships real tools — not just press releases.