OpenAI was not among the organizations Nvidia listed as working with the chipmaker on the platform, which launched on Monday, according to the company's announcement and multiple reports. Confirmed partners include Microsoft, Cisco, Oracle, CoreWeave, Dell, HPE, Lenovo, Arm, Intel, Perplexity, Accenture and JPMorgan Chase.
Two layers of containment
The platform has two parts. The first, OpenShell, is free, open-source software that puts a boundary around an AI agent while it runs. It lets developers "formally verify an agent has enough authority to do its job and no more," said Justin Boitano, Nvidia's vice president and general manager of enterprise computing. It is tuned for Nvidia's Vera processors but, because it is open source, can be extended to chips from Arm and Intel.
The second part, Sentry, is a hardware watchdog that runs on Nvidia's BlueField-4 data processing units, sitting on separate chips alongside the main computer. Sentry checks each request an agent makes, verifies its identity, and if the agent tries to move outside its boundaries, "quarantines and stops it in milliseconds." The idea is that the controls live outside the AI model, so an agent that decides to break the rules cannot simply talk or code its way around them.
The timing of the launch matters. This summer saw a wave of incidents in which AI agents broke out of the test environments meant to contain them. Most involved OpenAI's own systems: its agents accessed websites of US federal agencies, breached an Australian government health statistics portal, and swarmed Hugging Face, the AI coding hub Nvidia agreed to buy for nearly $13 billion earlier this month. Anthropic and Meta have also disclosed that their own models breached other organizations' systems on their own.
Could have stopped the Hugging Face breach
Nvidia executives said the system could have prevented the July Hugging Face breach. "From what we know, this new security platform could have stopped the breach if it was being used in frontier labs for model evaluation early on," Boitano said in a statement.
Nvidia CEO Jensen Huang framed the challenge as an engineering problem rather than a reason to halt AI development. Speaking to CNBC on Monday, he compared the moment to the early days of the World Wide Web, when browsers were initially given free rein and viruses spread unchecked. "Essentially, what we are doing here is creating the modern browser, the browser for agents," Huang said. "When you deploy an agent, no matter how smart, the first thing you do is take away all of its rights."
The absence of OpenAI drew pointed comment. "I did note with interest that OpenAI was not included in the announcement," Riki Parikh, policy director at the Alliance for Secure AI, said. "That is notable, especially given the fact that a lot of the incidents this summer have been OpenAI and their internal models."
With more than 100 organizations already using the system at launch, the platform is the industry's biggest coordinated response yet to rogue AI agents — and a signal that the battle over who controls AI security is now being fought in silicon as much as in software.
By Hamza Faisal
