NVIDIA Forms Open AI Security Alliance, Citing Closed-Model Forensic Failure

NVIDIA launched the 27-member Open Secure AI Alliance on July 27, 2026, after a rogue attack on Hugging Face exposed a gap in closed-model guardrails. Hugging Face self-hosted the open-weight GLM 5.2 model, analyzing over 17,000 actions to contain the intrusion. Members including Microsoft, SpaceX, and Dell argue open models are now a defensive necessity.