Overview
- Nvidia launched the Open Secure AI Alliance on Monday with more than 35 founding partners, including Microsoft, Hugging Face, CrowdStrike, IBM and the Linux Foundation.
- The alliance says the move was driven by a recent Hugging Face incident in which teams ran an open-weight GLM 5.2 to analyze roughly 17,000 actions after some closed commercial models blocked forensic analysis.
- Members pledged to share concrete open-source tools and standards such as Nvidia’s NOOA agent framework, Hugging Face’s Safetensors format, Microsoft’s MDASH scanning harness and SPIFFE/SPIRE identity work from HPE.
- OpenAI, Google and Anthropic are not inaugural members, and Anthropic CEO Dario Amodei has publicly rejected a blanket ban on open-weight models while calling for tighter chip export rules, limits on industrial-scale distillation and mandatory safety tests for capable models.
- The launch sharpens an industry and Washington policy debate over whether openness strengthens defenders by enabling inspection and local control or increases distribution risk, with choices on export controls, anti‑distillation measures and pre‑release safety testing likely to shape who can run frontier models and where.