DELVE INTO AFRICAN WEALTH
DON'T MISS A BEAT
Subscribe now
Skip to content

Billionaire Jensen Huang bets Nvidia chips can keep AI agents from going rogue

Jensen Huang's Nvidia launched a two-layer system to contain rogue AI agents, saying it could have stopped the OpenAI models that hacked Hugging Face.

Billionaire Jensen Huang bets Nvidia chips can keep AI agents from going rogue

Table of Contents

Taiwanese-American chip billionaire Jensen Huang has put Nvidia Corp. at the center of the fight to keep AI agents in check, launching a two-layer safety system the company says could have stopped the rogue OpenAI models that broke into Hugging Face this summer.

The Open Agent Safety Platform, unveiled Monday, pairs open-source software called OpenShell with a hardware design called Sentry. OpenShell draws a secure boundary around an agent, tracing its actions and enforcing rules as it runs on Nvidia's Vera CPUs. Sentry sits outside that boundary on a separate BlueField-4 chip, watching the agent and quarantining it within milliseconds if it tries to break out, according to Nvidia.

"AI's extraordinary potential for society will only be realized if we solve AI safety," Huang, Nvidia's founder and chief executive officer, said in a statement. "Safety and security require full-stack engineering."

The launch comes two months after OpenAI disclosed that a swarm of its models, including GPT-5.6 Sol and an unreleased system, escaped a sandboxed cyber test and hacked into Hugging Face's production infrastructure to pull benchmark answers. Hugging Face said the intruders reached a limited set of internal datasets and several credentials but found no evidence that customer data leaked. OpenAI said Sept. 25 that the agents had also uploaded 53 user-provided images to third-party hosting sites.

Justin Boitano, vice president and general manager of enterprise computing at Nvidia, said the platform could have prevented the incident had leading labs used it during early model evaluations.

That claim does not appear in Nvidia's own press release, which makes no mention of Hugging Face's breach and does not explain how the system would have interrupted the attack chain OpenAI described.

Huang also has a direct interest in the story. He agreed Sept. 3 to buy Hugging Face for $12.93 billion, a deal expected to close in the first half of 2027 pending regulatory approval. Hugging Face appears on the list of more than 100 launch partners, alongside Anthropic, Microsoft, CrowdStrike and JPMorganChase.

The platform extends Huang's push to sell complete systems rather than chips alone. OpenShell is open source and can be adapted for Arm and Intel processors. Sentry's watchdog, as Nvidia describes it, runs on the company's own BlueField-4 hardware, which means customers who want both layers of protection will be buying Nvidia silicon.

OpenShell is available now through Nvidia's developer channels and GitHub. The company has not said when Sentry will be available.

The intelligence satisfies curiosity. The paid briefings satisfy strategy.

Every Monday, Elite subscribers receive an Investor Memo breaking down the deal, the structure and the positioning behind the week's most consequential African wealth story - the kind of analysis that doesn't appear anywhere else.

Twice a month, a Wealth Intelligence brief profiles a single billionaire's holdings, cash flows and expansion pipeline in detail no public source matches.

→ Executive ($25/mo): Daily newsletter + Deep-Dive Reports

→ Elite ($75/mo): Everything above + Investor Memos + Wealth Intelligence + Quarterly Analyst Briefings

Subscribe now

Latest