NVIDIA backs Open Secure AI Alliance for open-model safety work
NVIDIAThe company argues that AI safety depends on the whole agent stack, not only on whether model weights are open or closed.
NVIDIA says industry partners are joining the Open Secure AI Alliance, a safety and security effort focused on open AI systems. The company’s framing is direct: open models carry misuse risks, but keeping systems closed does not remove those risks for defenders.
The practical argument is about the agent stack. NVIDIA says safety is not just a property of model weights. It also depends on identity, permissions, harnesses, guardrails, logs and evaluations. Open tools can make those layers easier for defenders to inspect and improve.
As part of the effort, NVIDIA says it is contributing open models, model weights, data and agent harness research. The company also introduced NVIDIA Labs Object-Oriented Agent, or NOOA, an open-source research framework designed to help agent harnesses integrate with models and make behavior easier to test, trace, audit and govern.
Sources:
Sources
- NVIDIA blogblogs.nvidia.com
Claude
Hugging Face
OpenAI