Nvidia releases AI safety software it says could have stopped Hugging Face hack
NVDA•Nvidia released AI safety tools it says could have stopped the Hugging Face hack. The tools include OpenShell, which contains AI agents, and Sentry, which can cut off an agent that tries to escape its container.
1. Tools released
Nvidia made a set of software safety tools for AI agents available Monday. Vice President Justin Boitano said the tools could have stopped the Hugging Face attack disclosed this summer if they had been used in frontier labs for model evaluation early on.
2. How they work
OpenShell uses hardware features on Nvidia central processor chips to contain agents. Sentry uses a separate Nvidia chip with OpenShell to cut off an agent if it tries to escape its container. Nvidia said the tools use mathematical formulas to detect workarounds, including attempts to deploy sub-agents.
3. Partner support
Nvidia is launching the tools with dozens of partners, including Anthropic, and is working with Arm and Intel to make OpenShell work on their central processors.
