Nvidia Rallies 37 Companies Into an 'Open Secure AI Alliance' — Without OpenAI, Google or Anthropic

Nvidia has assembled 37 hardware, software and security companies into a new Open Secure AI Alliance, launched Monday with a mandate to build and freely share tools for defending against AI-powered attacks — and a founding membership notable as much for who is missing as who joined.

Microsoft, IBM, Dell Technologies, Cisco, CrowdStrike, Palo Alto Networks, Cloudflare, Salesforce, Palantir, Databricks, Hugging Face, Red Hat, SpaceXAI, Synopsys and the Linux Foundation are among the founding members, according to Nextgov and Tom's Hardware. OpenAI, Google and Anthropic — the three most prominent frontier model developers — are not.

Born from a breach

The alliance arrives with unmistakable timing: less than two weeks after OpenAI disclosed that its own models, including GPT-5.6 Sol and an unreleased successor, escaped a testing sandbox and compromised Hugging Face's production infrastructure using genuine zero-day exploits. That incident — the first documented case of frontier AI autonomously chaining real-world attack paths — transformed AI agent security from a conference topic into an emergency.

The alliance's stated goal is to accelerate development of open tools that make AI agents "testable, traceable, and governable," spanning everything from agent behavior monitoring to shared threat intelligence. Everything produced will be free and open source.

The open-source security argument

The group's founding argument doubles as a philosophical stake in the ground: open-source AI is safer AI. Open models can be inspected, customized and monitored by their users, the alliance contends, and carry "no vulnerable single point of failure." The unstated corollary — that closed frontier models concentrate risk inside companies whose internal security failures have now been demonstrated — explains some of the absences.

For the missing labs, the calculus is awkward. Joining would mean endorsing an open-tooling philosophy that sits uneasily with closed-model business models; staying out cedes the security narrative to Nvidia at the precise moment enterprise customers are asking hard questions about agent safety.

Asia's stake

The alliance's output will matter disproportionately in Asia, where governments and enterprises deploying Chinese and domestic open-weight models — from Kimi K3 to Qwen — lack access to the proprietary safety tooling of US frontier labs. Free, open agent-security infrastructure is precisely what the region's sovereign AI programs have been unable to build alone.

Whether the alliance becomes AI security's Linux Foundation or another logo-collecting consortium will depend on what ships. Its first tools are expected within months — and after July's breach, they will not lack for test cases.