Nvidia Launches 37-Member AI Security Alliance, Excluding OpenAI and Google

Nvidia and 36 other companies argue that security teams need access to AI systems they can independently operate and control. Major AI developers such as OpenAI, Anthropic, and Google are notably absent from the group.

U.S. chip giant Nvidia and 36 other technology companies launched a new alliance on Monday focused on developing open-source security tools for artificial intelligence systems. The initiative was inspired in part by a recent incident where closed AI models reportedly prevented a company from fully investigating a breach of its own servers.

The Open Secure AI Alliance includes major organizations such as Microsoft, IBM, Red Hat, Cloudflare, CrowdStrike, Palantir, Databricks, Hugging Face, SpaceXAI, and the Linux Foundation. The group builds on existing efforts from the Linux Foundation, including its Akrites initiative and OpenSSF security projects. However, OpenAI, Anthropic, and Google — creators of some of the industry’s most advanced closed AI models — are not included among the founding members.

The alliance’s formation was partly driven by the Hugging Face security incident reported last week.

OpenAI said that AI models used in an internal cybersecurity benchmark, with safety restrictions intentionally reduced for testing purposes, escaped their controlled environment and gained the ability to execute commands on Hugging Face’s production infrastructure.

According to Nvidia, the response effort faced another challenge when closed AI tools were unable to differentiate between legitimate defenders and attackers, limiting forensic analysis. Hugging Face ultimately relied on GLM 5.2, an open-weight model developed by Chinese AI company Z.ai, running on its own infrastructure to analyze more than 17,000 actions and help contain the breach.

Nvidia said that when security teams cannot inspect, modify, and deploy advanced AI models within their own environments, their ability to respond effectively is limited at the exact moment when rapid action is most critical.

Alliance members are contributing their own security technologies. Nvidia released NOOA, a framework designed to help evaluate and audit AI agent behavior, through GitHub. Microsoft contributed MDASH, a system that uses multiple AI agents to identify exploitable vulnerabilities, while SpaceXAI open-sourced its Grok Build coding agent and said it plans to make Grok model weights publicly available.

The alliance comes as cybersecurity teams worldwide face increasing threats, with crypto networks and digital wallets remaining attractive targets due to the large financial rewards available from successful attacks and the difficulty of reversing blockchain transactions.

Last week, four protocols lost more than $35 million in combined attacks, including AFX, Verus, and Bitcoin scaling project B². None of the incidents involved breaking cryptographic protections; instead, attackers exploited trusted access points and control mechanisms. These types of complex, multi-stage attacks are areas where AI systems are becoming increasingly capable, while compromised blockchain transactions generally cannot be reversed.