NVIDIA and Microsoft Form Open Secure AI Alliance With 27 Founding Members

NVIDIA created the Open Secure AI Alliance, a consortium of 27 companies focused on enhancing cybersecurity during the age of AI. This includes Microsoft, SpaceX, Dell, HPE, IBM, Red Hat, Hugging Face, and the Linux Foundation. However, OpenAI, Google, and Anthropic are absent from this group.

They aim to “identify and remediate vulnerabilities through open technologies” by building upon efforts made by the Linux Foundation and OpenSSF. NVIDIA will share open models, weights, and agent tools while HPE will provide cryptographic verification standards for AI agents. Hugging Face will contribute its Safetensors format for securely storing model weights.

They use an example from Hugging Face where an AI agent hacked their system and performed over 17,000 autonomous actions within it. Commercial security products were unable to detect this because of inherent constraints, but an open source model called GLM 5.2 was able to analyze everything the attacker did and prevent further damage.

They plan to advocate for open models being considered an important part of cybersecurity instead of something dangerous and fight against blanket bans on open AI systems.