NVIDIA launched the Open Agent Safety Platform, an open software platform and reference system that it says is designed to enhance and strengthen AI security, from agent testing to deployment. The timing of this deployment comes at a time when AI security is becoming a growing concern.
Chief among these concerns is the fear of rogue AI agents acting independently or working as a collective in order to hack agencies. Such is the case of the Hugging Face scare that occurred this year.
Gatekeeping AI Agents
“AI’s extraordinary potential for society will only be realised if we solve AI safety,” Jensen Huang, founder and CEO of NVIDIA, said. “As we continue to discover the frontier of AI capabilities, we must accelerate discovery at the frontier of AI safety. Safety and security require full-stack engineering. NVIDIA Open Agent Safety Platform brings together industry, researchers and public-sector organisations to share best practices, align on evaluation methods and foster international cooperation. Together, we can raise the bar for global AI safety.”
In broad strokes, NVIDIA’s Open Agent Safety Platform will enable full-stack governance that will oversee the safety and efficacy of AI agents. First, there’s NVIDIA OpenShell, the brand’s own secure runtime software that sets boundaries for agents running on CPUs, and specifically, through Vera, its first purpose-built CPU for agentic AI. That said, OpenShell can also be made to work with other third-party platforms, including those from ARM and Intel.
The Open Agent Safety Platform also includes NVIDIA Sentry, which the company describes as an “out-of-band” watchdog, running on its BlueField-4 DPU and continuously monitoring agent behaviour. Basically, if the platform determines the AI agent to be acting out of bounds or if it is deemed to be a threat, it flags it down and isolates it.
Sentry is, in turn, powered by NVIDIA DOCA, software that the company says provides the watchdog with the programmable capabilities to inspect agent requests and responses, and basically make all the decisions to determine if an agent is, in layman’s terms, friend or foe.
As of this publication, NVIDIA says that more than 100 companies have begun using its Open Security Safety platform. That includes, among others, Accenture, Armadin, Cadence, Cognition, CrowdStrike, Cisco, Dassault Systèmes, Deloitte, EY, Hugging Face, IBM, Irregular, Perplexity, Microsoft, SAP, Scale AI, ServiceNow, Siemens, Synopsys, OpenClaw, Palantir and Palo Alto Networks.
The Open Agent Safety Platform, including OpenShell and skills, are already available through NVIDIA’s developer resources page and GitHub.
