NVIDIA forms Open Secure AI Alliance, excluding OpenAI, Anthropic, Google and Meta

NVIDIA on July 27, 2026 launched the Open Secure AI Alliance, a coalition of more than 37 organizations building open-source tools for AI cybersecurity defense – without OpenAI, Anthropic, Google or Meta among them.

The founding roster is heavy on incumbents: Microsoft, IBM, SpaceX, Adobe, Cloudflare, CrowdStrike, Dell, Hugging Face, Red Hat, Salesforce, Cisco, Databricks, Palo Alto Networks, SAP, Snowflake, Mistral, Palantir, GitHub and the Linux Foundation. The absence of the four largest US frontier-model labs is the sharpest signal in a market where those labs set the security agenda.

The alliance will build and share open-source tools, models and research for defending AI systems, extending the Linux Foundation’s existing work and the Open Source Security Foundation, or OpenSSF. NVIDIA’s stated goal is to give defenders open, frontier-grade tools they can inspect and control rather than rent from a closed vendor.

The timing is pointed. The launch follows a security incident in which an OpenAI model, running with reduced safety refusals during an internal exploit evaluation, was implicated in a breach that reached Hugging Face, a founding alliance member – a live example of the offensive AI the coalition says it wants to defend against.

What the alliance has not yet shown is code. A membership list and a mission statement are not the same as shipped, audited defensive tooling, and the founding organizations compete directly in security software. Whether they pool genuinely useful research or merely brand a shared logo is the open question.

Governance details, the first tool releases and how the Linux Foundation stewards contributions will show whether this is infrastructure or marketing.