NVIDIA forms Open Secure AI Alliance without OpenAI or Google

The interesting part of NVIDIA's new Open Secure AI Alliance is not the size of the roster, it is who declined to show up. OpenAI, Anthropic and Google are absent from a coalition whose founding thesis is that closed frontier AI on its own cannot defend cyberspace.

Announced on July 27, 2026, the alliance brings NVIDIA together with Microsoft, IBM, Red Hat, the Linux Foundation, Hugging Face, Cisco, CrowdStrike, Dell, HPE and Palo Alto Networks, among others. Jensen Huang put the pitch in a sentence: "Attackers have frontier AI. Defenders need a frontier AI ecosystem, the best open and closed models, force-multiplied by a global community." Reporting differs on the exact headcount, so take the specifics as reported, not settled; one account cites 27 founding members, another more than 40.

The stated trigger is a recent cyber incident at Hugging Face. According to Help Net Security, defenders turned to the open-weight GLM 5.2 model on their own infrastructure to analyze more than 17,000 actions during the response. The alliance's framing is that when defenders are limited to closed commercial models, their response envelope narrows in exactly the moments it needs to widen.

The honest caveat is that a launch document with a member list is not a shipped set of tools. What the reporting does not give you is governance, funding, or which specific defensive artifacts each member has actually committed to release, and whether the absent frontier labs plan to engage on their own terms. Still, the underlying play is worth watching: a big-tech coalition arguing to regulators that open-weight models are defensive infrastructure hands smaller security teams and researchers a seat at a table that lately belonged to a handful of closed vendors.