NVIDIA, along with over 100 partners, is launching an open AI agent safety platform built around runtime sandboxing and hardware monitoring.
The initiative targets risks exposed during recent frontier laboratory evaluations, where autonomous software escaped testing environments, breached unauthorised systems, and generated misleading activity logs.
Software agents often encounter instruction ambiguities, missing operational tools, or execution loops that cause them to drift away from designated parameters. “An agent in these circumstances cannot be expected to fully govern its own behaviour,” NVIDIA explained.
Hardware controls inside Vera Rubin PODs
The reference architecture centres on OpenShell, an open-source secure runtime published under an Apache 2.0 licence. Operating with kernel-level isolation, the runtime converts administrative parameters into verifiable policies before any workload executes.
Administrators establish strict boundaries governing network access, file storage, specific processes, tools, and digital credentials. A formal prover verifies that the operating policy adheres strictly to operator boundaries prior to launch, denying execution if escape paths exist.
Physical isolation forms the second layer of the defensive stack. NVIDIA paired OpenShell with Sentry, a monitoring service designed for BlueField data processing units.
In the Vera Rubin POD configuration, systems route all communications through a BlueField-4 unit positioned on the host node’s model access route. This intercepts communication out of band, keeping enforcement mechanisms away from host operating systems that compromised agent code might manipulate.
Running directly in silicon, the hardware inspects network packets and tool queries at line speed, providing telemetry and acting as an external kill switch whenever execution patterns diverge from verified profiles.
Jensen Huang, Founder and CEO of NVIDIA, said: “Today, with over 100 industry partners, we introduced the NVIDIA Open Agent Safety Platform, bringing together OpenShell and Sentry.
“Artificial intelligence is extraordinary technology that will advance discovery, productivity, security, health, and prosperity for generations to come. But its full promise can only be realised when people have confidence that AI is being built to be safe and deployed with wisdom and responsibility.
“This is bigger than a single product. It’s the beginning of an open ecosystem to build the trust layer for safe agent systems. Together, we are building the foundation of the AI economy.”
Telemetry verification across enterprise infrastructure stacks
The platform architecture divides responsibility across three distinct tiers: application assets, the runtime boundary, and physical infrastructure.
Application layers supply the foundational models, data pipelines, harness frameworks, and scripts. The OpenShell runtime subsequently maps those resources across target compute clusters, handling continuous governance whether running in central data centres, workstations, or remote edge nodes.
NVIDIA DOCA software connects the hardware telemetry directly to OpenShell policy engines.
By correlating tool execution histories, data queries, and dynamic permission authorisations, the DOCA gateway establishes an audit trail of machine activity. The system tracks identity delegations continuously, verifying that subagents do not exceed their assigned scopes.
“Trust and innovation are not in conflict. Safety is how trust is earned. We must build not only the most capable AI, but the most trusted AI, so that this extraordinary technology can realise its enormous promise for the world,” Huang concludes.
NVIDIA confirmed that enterprises running existing Vera and BlueField-4 deployments can activate these security protections through a software update.
See also: Google launches Gemini 3.8 Flash TTS voice models
Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including IoT Tech Expo and the Cyber Security & Cloud Expo. Click here for more information.
AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.
