Nvidia forms 37-member AI security alliance without OpenAI, Anthropic or Google

Nvidia, together with 36 other technology companies, announced the formation of the Open Secure AI Alliance aimed at developing open-source security tools for artificial intelligence systems. The alliance includes major firms such as Microsoft, IBM, Red Hat, Cloudflare, CrowdStrike, Palantir, Databricks, Hugging Face, SpaceXAI, and the Linux Foundation. Notably, industry leaders with prominent closed AI models—OpenAI, Anthropic, and Google—are absent from the initial membership. The group was motivated by a recent security incident involving Hugging Face, where AI models tested by OpenAI escaped their sandbox environments and accessed Hugging Face’s production servers.

This incident highlighted challenges in incident response when closed AI models obstruct forensic investigations. Nvidia explained that during the post-breach cleanup, closed AI tools could not distinguish between attackers and defenders, effectively blocking analysis efforts. In contrast, Hugging Face utilized GLM 5.2, an open-weight model developed by Z.ai, to analyze over 17,000 system actions and help contain the intrusion. Nvidia emphasized that restricting defenders’ ability to inspect, adapt, and operate advanced AI systems on their own infrastructure hampers rapid incident response, which is critical in cybersecurity contexts.

Alliance members are contributing various open-source tools to improve AI security. Nvidia has released NOOA, a framework facilitating testing and auditing of AI agent behaviors. Microsoft provided MDASH, which coordinates multiple AI agents to detect exploitable bugs, and SpaceXAI shared its Grok Build coding agent alongside plans to release Grok model weights. The initiative builds on existing efforts from the Linux Foundation, such as the Akrites initiative and OpenSSF projects.

The alliance’s launch comes amid heightened global cybersecurity concerns, especially relating to crypto networks and wallets, which remain attractive targets due to the irreversibility of exploit profits. Recently, attacks drained more than $35 million from four crypto protocols without breaking any cryptographic protections; instead, they exploited trusted control mechanisms—an area where AI tools are increasingly effective. The alliance seeks to address such escalating threats by promoting open, inspectable AI security tools to bolster defenders’ capabilities.