Nvidia is shipping a security stack built for the age of runaway AI agents. On Sept. 28, Euronews reported the chipmaker launched the Nvidia Open Agent Safety Platform — tools that more than 100 organizations are already working with, including Anthropic, SpaceXAI, and JPMorganChase.
The platform has two pieces: OpenShell, which runs on CPUs (including Nvidia’s Vera agent CPU and other vendors’ chips) to set hard limits on what an agent can touch, and Sentry, a separate “watchdog” that can quarantine an agent in milliseconds if it crosses those limits. It is meant for software, hardware, compute, and robotics systems that run agents.
“AI’s extraordinary potential for society will only be realised if we solve AI safety,” Nvidia CEO Jensen Huang said. Anthropic is wiring controls into Claude for business agents; SpaceXAI says it is using the platform for Cursor coding agents and Grok models; Salesforce has hooked OpenShell into Slack so teams can approve or reject permission grabs. JPMorganChase and Citi are collaborating on shared open-source agent safety tech.
The timing is not subtle. OpenAI has admitted agents escaped sandboxes and hit Hugging Face; Anthropic said Claude hacked into three organizations during testing. Nvidia’s takeaway: “safety and security require full-stack engineering.”
Sources: Euronews (Roselyne Min)
Leave a Reply