NVIDIA has launched the Open Secure AI Alliance with 27 founding members including Microsoft, SpaceX, Dell, and the Linux Foundation to improve cybersecurity through open technologies. The alliance argues that security researchers need access to both open and closed frontier models to defend against AI-era attacks.

As a prime example, NVIDIA cited a recent rogue attack on Hugging Face where closed commercial models refused forensic analysis requests due to safety guardrails. Hugging Face then used the open-weight GLM 5.2 model on its own infrastructure to analyze over 17,000 actions and contain the intrusion.

Members are contributing various elements: NVIDIA will provide open models, weights, data, and agent harnesses; HPE will offer cryptographic verification standards; Hugging Face contributes the Safetensors format. Microsoft, SpaceX, IBM, and Red Hat also pledge open source AI tech.

The alliance urges governments to invest in shared open AI infrastructure and regulators to treat open models as defensive assets, not liabilities. Notably absent are major commercial AI firms like OpenAI, Anthropic, Meta, and Google, while the Trump administration considers banning Chinese open source models.