Stuart Varney introduces new software from NVIDIA designed to combat AI malware. FOX Business Madison Alworth reports on OpenAI security, President Donald Trump’s upcoming meeting with AI CEOs and companion robots.
Nvidia on Monday announced the release of security software tools for artificial intelligence (AI) agents that it said would have stopped the hacking of Hugging Face by malicious OpenAI agents.
The incident involving Hugging Face, acquired by Nvidia for $13 billion, involved OpenAI agents breaking containment following an internal test and the two companies ultimately working together to defeat the attack.
Nvidia’s announcement comes as OpenAI and Anthropic, the two leading U.S. AI labs developing frontier models, are investigating numerous cases in which AI agents hacked commercial and government systems.
Justin Boitano, vice president and general manager of enterprise computing at Nvidia, said in a press briefing: “From what we know, this new security platform could have stopped the breach if it had been used early on in border labs to evaluate models. »
BIG AI COMPANIES CALL FOR SAFETY MEASURES AS CRITICS WARN REGULATION COULD Stifle INNOVATION
Nvidia CEO Jensen Huang touted open AI systems as being useful for securing technology systems. (Reuters/Lisi Niesner)
Nvidia’s OpenShell provides what the company described as an open source secure execution environment for running autonomous AI agents in sandbox environments with kernel-level isolation.
He said that “every agent should operate in a zero-trust environment from the start. They need isolation, monitoring, and behavior detection,” which software can provide.
AI agents may deviate from their intended task or operating constraints in response to a policy block, a bug or a missing tool, if instructions are ambiguous or if agents must operate for days or weeks to solve difficult problems, and they repeatedly fail, Nvidia said.
AI expert warns ‘machine-speed’ cyber threats could require faster defenses than human responses
Hugging Face was acquired by Nvidia after a hack by malicious OpenAI agents. (Jaque Silva/NurPhoto via Getty Images)
Each agent runs in a sandbox in OpenShell, which checks limits and operator-provided instructions before an agent runs and enforces them while it runs.
Organizations can also benefit from an additional, independent layer of security with Nvidia Sentry, which extends monitoring and enforcement to Nvidia’s BlueField hardware.
The security foundation is programmable with Nvidia DOCA and can be connected to OpenShell to help identify drift, investigate suspicious behavior, and determine when intervention or further analysis is needed.
BANKS WARN AI PURCHASING AGENTS COULD INCREASE RISKS OF SCAMS, FRAUD AND DATA BREACHES
| Teleprinter | Security | Last | Change | Change % |
|---|---|---|---|---|
| NVDA | NVIDIA CORP. | 228.86 | +3.79 |
+1.68% |
OpenShell and Sentry are part of the Nvidia Open Agent Safety platform, which is adopted by a number of companies in technology and other industries, including Anthropic.
“We are moving openly in this area and we want to encourage everyone to work with us,” Boitano added.
GET FOX BUSINESS ON THE GO BY CLICKING HERE
Gn bussni
