Nvidia (NVDA) is expanding its efforts to prevent AI agents from going rogue and partnering with Agile Robots on a new humanoid robot reference design.
The company unveiled its OpenShell reference architecture at its GTC Berlin event in Germany on Wednesday, saying it will allow customers to create, contain and observe AI agents.
The reference architecture is a basic design for a server or other platform that customers and partners can customize to their own needs.
The setup allows customers to run AI models on Nvidia’s Vera Rubin system, with the agent software running on the company’s Vera processors and Nvidia’s BlueField-4 data processing units acting as security guards that block agents if they attempt unauthorized actions.
218.64 +2.16 (+1.00%)
At 10:29:41 a.m. EDT. Open market.
The company says it is working with CoreWeave (CRWV), Crusoe (CUES.PVT), Lambda (LAMD.PVT), Nebius (NBIS), Oracle Cloud Infrastructure (ORCL) and others to deploy the reference architecture.
“AI capabilities and security must advance together. As companies prepare to deploy billions of AI agents, security and control must be integrated into the IT architecture,” Nvidia CEO Jensen Huang said in a statement.
“Nvidia created OpenShell to give agents the freedom to do meaningful work within clearly defined boundaries – contained, monitored and under control,” he said. “With the OpenShell reference architecture, agents can think on Rubin, act on Vera, and be governed on BlueField.”
Nvidia released its OpenShell and Nvidia Sentry software late last month, claiming they could have prevented incidents such as OpenAI’s (OPAI.PVT) Hugging Face hack.
Learn more: FTC reportedly looking into whether OpenAI and Anthropic violated consumer protection laws
The AI industry continues to grapple with summer revelations that agents from major AI labs escaped their secure testing environments and reached the open Internet, where they hacked various websites and services.
Anthropic (ANTH.PVT), Google (GOOG, GOOGL) and Meta (META) also said their AI agents had been released.
Security experts say the hacks were largely the result of poor security posture on the part of AI labs and their testing partners.
According to Nvidia, the OpenShell reference architecture can quarantine agents that attempt to go malicious and stop them in “milliseconds.”
On the robotics side, Nvidia Agile Robots has developed a reference humanoid robot that works on its Nvidia Isaac GR00T software platform and its Nvidia Jetson Thor computer.
Agile One SN from Agile Robotics is a 4.59-foot-tall humanoid robot designed to help developers build general-purpose robot “brains” to accomplish a wide range of tasks.
Gn bussni
