NVIDIA Unveiled a Platform for Monitoring AI Agents

• NVIDIA has launched the Open Agent Safety Platform to monitor autonomous AI agents.
• The system can isolate an agent within milliseconds if it goes beyond predefined boundaries.
• More than 100 organizations have joined the initiative, including Anthropic, Microsoft, JPMorganChase, and Salesforce.

NVIDIA unveiled the Open Agent Safety Platform — an open platform for monitoring and securing autonomous AI agents during testing and in production. It is designed to constrain what these systems can do and stop them if they attempt to operate outside permitted limits.

The platform includes two core components. OpenShell creates a secure runtime environment and controls an agent’s access to data, tools, and the network. 

NVIDIA Sentry acts as an independent watchdog on BlueField-4 DPU processors and can isolate an agent that breaks the rules within milliseconds. According to the developers, this separate control layer is independent of the AI itself and cannot be disabled by the agent.

“Safety and security require comprehensive engineering,” said NVIDIA CEO Jensen Huang. He added that the new platform should help companies, researchers, and government agencies develop shared approaches to safe AI deployment.

OpenShell is open source and can run not only on NVIDIA infrastructure, but also on Arm and Intel platforms. More than 100 organizations are participating in the project, including Anthropic, Microsoft, JPMorganChase, Salesforce, Cisco, CrowdStrike, Palantir, and Hugging Face.

The launch comes amid a series of incidents involving AI agents escaping their test environments. OpenAI previously began developing an automatic “kill switch” for AI after a case in which its agents gained access to the internet and Hugging Face infrastructure during testing.

At Anthropic, the issue of remotely stopping models has also become a separate topic of debate. As part of the company’s dispute with the Pentagon, a US court found that after the models are deployed in national security systems, Anthropic has no backdoor access to them and cannot remotely modify or disable them.

At NVIDIA itself, they see such risks primarily as an engineering challenge. The company says that the combination of OpenShell software restrictions and Sentry’s independent hardware controls should make it possible to stop an agent’s dangerous actions even in cases where it tries to bypass protections at the application level.

Сообщение NVIDIA Unveiled a Platform for Monitoring AI Agents появились сначала на INCRYPTED.