Nvidia, the chipmaker whose processors run most of the world's artificial intelligence, has pulled more than 30 of the biggest names in tech into a new group with a pointed mission and an even more pointed guest list. Called the Open Secure AI Alliance, the coalition brings together Nvidia, Microsoft, SpaceX, Dell, IBM, Red Hat, HPE, Hugging Face and the Linux Foundation, among others, to build and freely share open-source tools for defending AI systems from attack. The idea is that cyber defenders need open models, the kind whose inner workings are public, so protection is not locked inside a handful of companies.
A breach that made the case The alliance did not appear out of nowhere. It was galvanised by a security incident this month at Hugging Face, a widely used AI platform, when the company first reached for closed commercial AI tools to investigate a hack, only for those tools to hit their own safety guardrails and refuse to help with the forensic work. Hugging Face then fell back on an open-weight model, GLM 5.2 from Beijing-based Z.ai, running it on its own systems to comb through more than 17,000 actions and shut the intrusion down. For the alliance's founders, that was the whole argument in miniature: when the closed tools froze, an open one did the job. Members are now each chipping in, with Nvidia releasing its own security-agent software on GitHub and Microsoft, IBM and Red Hat adding tools of their own.
Who skipped the party The most striking thing about the alliance is who is missing. OpenAI, Google and Anthropic, the three leading makers of closed AI models, are all absent, and the group is openly lobbying regulators to treat open models as "defensive assets, not liabilities" rather than security threats. That message is landing at a delicate moment, with the Trump administration reportedly weighing a wide ban on Chinese open-weight AI models, the very category that helped Hugging Face contain its breach. The founding statement puts the philosophy bluntly: "The right response is not to deny defenders access to capable open systems."
The real test now is whether an open-source coalition can set the terms of AI security, or whether the big closed-model players and policymakers push back, and that is the fight just beginning.
