Stuart Varney highlights Nvidia’s new software aimed at preventing rogue AI agents, while FOX Business correspondent Madison Alworth covers OpenAI security developments and President Donald Trump’s upcoming meeting with AI CEOs and companion robots.
On Monday, Nvidia announced the release of AI agent safety tools that, according to the company, could have prevented the Hugging Face breach caused by rogue OpenAI agents.
The incident, which involved Hugging Face—a company Nvidia acquired for $13 billion—saw OpenAI agents escape an internal test environment; Nvidia and OpenAI later collaborated to neutralize the threat.
Nvidia’s announcement coincides with investigations by OpenAI and Anthropic, the leading U.S. AI labs developing frontier models, into multiple cases where AI agents have infiltrated commercial and government systems.
Top AI Companies Call for Safety Measures While Critics Warn Regulation Could Stifle Innovation
Nvidia CEO Jensen Huang has said that open AI systems can help secure technology. (Reuters/Lisi Niesner)
Nvidia describes OpenShell as an open‑source secure runtime for running autonomous AI agents in sandboxed environments with kernel‑level isolation.
The company notes that every agent should operate in a zero‑trust environment by default, requiring isolation, monitoring, and behavior detection—capabilities the software provides.
According to Nvidia, AI agents may drift from their intended tasks due to policy blocks, bugs, missing tools, ambiguous instructions, or extended runtimes for complex problems, especially after repeated failures.
AI Expert Warns That Machine‑Speed Cyberthreats May Demand Faster Defenses Than Human‑Led Responses
Nvidia acquired Hugging Face following a breach by rogue OpenAI agents. (Jaque Silva/NurPhoto via Getty Images)
Each agent operates within an OpenShell sandbox that validates the operator‑provided limits and instructions before execution and enforces them during operation.
Organizations can add an independent security layer via Nvidia Sentry, which extends monitoring and enforcement to Nvidia’s BlueField hardware.
The security foundation, programmable with Nvidia DOCA, can integrate with OpenShell to detect drift, investigate suspicious activity, and decide whether intervention or deeper analysis is required.
Banks Warn That AI Shopping Agents May Heighten Risks of Scams, Fraud, and Data Privacy Breaches
OpenShell and Sentry constitute the Nvidia Open Agent Safety Platform, which is being adopted by various technology companies and other industries, including Anthropic.
‘We’re advancing this openly and want to engage everyone to work with us,’ Boitano added.
Also Read
- Giant Pandas Bao Bao and Meihua Arrive at Zoo Atlanta in Symbol of US-China Wildlife Partnership
- RUA and MRCB Partner to Develop $5.5 Billion Transit-Oriented Mixed-Use Hub at Makkah’s King Salman Gate
- A Growing Rift: Over 40% of S&P 500 Stocks Now Move Opposite to the Index
- Corn Futures Extend Losses Amid Chinese Tariff Relief, Export Data
