Nvidia launches safeguards against rogue AI agents
Introduced on Monday, the Open Agent Safety Platform combines OpenShell, an open-source system that controls what agents can access and do, with a separate monitoring layer called Sentry. Nvidia says OpenShell denies access by default and enforces rules outside an agent’s own process. All decisions to allow or block an action can be audited.
Sentry independently monitors agents and can isolate them if they attempt to exceed those limits.
Huang said OpenShell gives agents “clear, enforceable boundaries” and traces their actions as they work. Sentry uses hardware-based monitoring and can contain or isolate suspicious agents within milliseconds, he added.
The launch follows incidents in which AI agents accessed systems without authorization, including a breach of the AI development platform Hugging Face involving OpenAI agents. Nvidia says the new safeguards could have prevented that breach if they had been used during earlier testing.
“Artificial intelligence is extraordinary technology that will advance discovery, productivity, security, health, and prosperity for generations to come,” Huang wrote. “But its full promise can only be realized when people have confidence that AI is being built to be safe and deployed with wisdom and responsibility.”
“Trust and innovation are not in conflict. Safety is how trust is earned,” he added.
Earlier, Qazinform News Agency reported that NVIDIA’s Board of Directors had authorized an additional $150 billion for the company’s share repurchase program, bringing the total remaining authorization to $235 billion.