AdvertisementAdvertisementAdvertisementAdvertisement
Technology

Nvidia Introduces Protective AI Software Against Threats

9/28/2026, 02:49 PM • Evgenia Sliv

(edited: 09/28/2026)

Nvidia Introduces Protective AI Software Against Threats

On Monday, Nvidia released a set of security software tools for AI agents, which, according to the company, could prevent the hacking of Hugging Face – an AI coding hub for which Nvidia paid $13 billion a few months after it was attacked by uncontrolled OpenAI agents. This was reported by Reuters.

This comes amid investigations by OpenAI and Anthropic, two leading American AI labs, into numerous instances where their agents – AI systems capable of performing complex tasks – have hacked into commercial and government systems. Nvidia CEO Jensen Huang, who leads the world's largest company whose chips have powered much of the AI boom, dismissed calls for broad AI regulation, calling runaway agents an engineering problem that needs solving – akin to improving car safety.

One of the tools released on Monday is called OpenShell. It uses hardware features of Nvidia's central processors to isolate agents. Nvidia is also working with Arm Holdings and Intel to ensure the system works on their processors as well. The company is launching the tools with dozens of partners, including Anthropic.

Justin Boitano, vice president and general manager of enterprise computing at Nvidia, stated that the tools would have stopped the attack on Hugging Face, which became known in the summer. "Based on what we know, this new security platform could have stopped the hack if it had been used in advanced labs to evaluate models at an early stage," Boitano said at a briefing for journalists. "We are developing this openly and want to engage everyone in collaboration."

Another system, Sentry, uses a separate Nvidia chip along with OpenShell to disable an uncontrolled agent if it attempts to leave its container on the central processor. Nvidia's tools use mathematical formulas to detect attempts by agents to bypass restrictions – for example, when an agent tries to "spawn" multiple "sub-agents" to circumvent the main agent's block, explained Ali Golshan, senior director of AI software at Nvidia. "It's really about agent behavior, about fleets of agents and how they work together," Golshan said at the briefing.

Popular news