Hajra Naz
A debate rages right now over recent rogue AI agent incidents. Are these incidents a step toward AGI? Or are they simply a more conventional engineering problem? Nvidia is offering its own answer to this question.
Nvidia CEO Jensen Huang introduced something new on Monday. He unveiled a toolkit of software and hardware products. These add independent security layers around AI agents. The goal is keeping agents within their test environments. This applies even if agents attempt to break out.
This release follows a string of hacking incidents recently. These involved AI models from Anthropic, Google, OpenAI, and Meta. In each case, models bypassed security controls. They escaped their testing environments and accessed real-world systems.
The first and most prominent example happened this summer. OpenAI agents breached Hugging Face while attempting a cybersecurity task. More incidents have followed since then. OpenAI even published a new site dedicated to reports of its AI agents going rogue.
Huang addressed this issue Monday during an interview with CNBC. He said Nvidia’s new Open Agent Safety Platform would have prevented these breaches.
Read More: We Can Handle AI Safety Ourselves and We Don’t Need AI Regulation, Says Nvidia’s Jensen Huang
Nvidia has made tens of billions of dollars selling GPU and CPU chips to AI labs. The company doesn’t support slowing down AI development. It also doesn’t support adding new industry regulations to solve this security problem.
Instead, Nvidia believes in a different approach. Some security controls should move outside the agent entirely. This creates a constant, independent security guard. That guard keeps AI agents in check at all times.
“AI’s extraordinary potential for society will only be realized if we solve AI safety,” Huang said in a statement. “As we continue to discover the frontier of AI capabilities, we must accelerate discovery at the frontier of AI safety. Safety and security require full-stack engineering.”
The new Nvidia Open Agent Safety Platform combines two key components. That includes OpenShell, Nvidia’s open source software. This controls what agents can access while operating. It also includes Sentry, an independent monitoring system.
Sentry runs on Nvidia’s BlueField-4 data processing units. According to Nvidia, placing Sentry on a separate processor matters. This differs from running it on the CPU or GPU where the AI agent actually operates. This separation provides an isolated view of the agent’s activity.
Read More: Jensen Huang Reveals Why Nvidia Could Surge 70% Next Year
OpenShell itself isn’t new. The company originally announced this software back in March. Still, Nvidia believes the combination of both tools matters most. Together, they provide the security layer needed to keep the industry moving forward safely.
OpenShell provides a software boundary around the agent. Sentry adds another line of defense at the hardware level. According to the company, this system continuously monitors behavior. It can “quarantine agents that attempt to move outside their boundaries in milliseconds.”
Nvidia listed dozens of companies supporting this effort. These companies plan to use the open source platform. That includes Anthropic, Arm, Microsoft, Oracle, and SpaceX. Notably, OpenAI is not listed as a participating company.
Huang shared additional context during his CNBC interview Monday. Work on this effort began about a year ago. It followed the introduction of OpenClaw. That’s an operating system of agents created by Peter Steinberger. In March, Nvidia released NemoClaw. This is an enterprise-grade AI agent platform. It represents Nvidia’s own version of OpenClaw, with security built directly in.
“When you deploy an agent, no matter how smart, the first thing you do is to take away all of its rights,” Huang said during his CNBC interview. He later compared these security measures to workplace management. Specifically, he compared it to how human employees, even executives, are managed within companies.
Nvidia’s release received wide support from a specific group. That includes people who have cautioned against slowing AI development. Their concern centers on China potentially surpassing the U.S. in AI capabilities.
Read More: Nvidia Confirms $12.93 Billion Deal to Buy Hugging Face
David Sacks weighed in on this announcement too. He’s a founder, venture capitalist, and former White House AI czar. He now co-chairs the President’s Council of Advisors on Science and Technology. He said Nvidia’s announcement serves as an important reminder. Agent safety is fundamentally an engineering problem, he argued.
“Recent breakouts weren’t proof that development must stop,” he wrote on X. “They were proof that the sandbox was too weak. The runtime environment was poorly designed and misconfigured.”
