NVIDIA has announced the NVIDIA Open Agent Safety Platform, an open software platform and reference system design for governing AI agents from testing through deployment. The announcement moves agent safety out of the application layer and into the runtime and the silicon underneath it.
How the NVIDIA Open Agent Safety Platform works
OpenShell is the software half: secure runtime software that traces every action an agent takes and enforces policy while it runs. NVIDIA says it is now broadly available and carries minimal overhead on Vera, its first CPU built specifically for agentic AI. The release describes OpenShell as open source; its GitHub repository confirms an Apache-2.0 licence, so it can be extended to work with third-party compute platforms including those from Arm and Intel.
Sentry is the hardware half, and it is the more interesting idea. It runs as an out-of-band watchdog on BlueField-4 DPUs, continuously monitoring agent behavior from an isolated trust domain. If an agent tries to move outside its software boundary, Sentry quarantines and stops it in milliseconds, enforcing policy in silicon rather than in code the agent can reach.
NVIDIA’s framing: the controls fail above the agent
NVIDIA attributes the platform to a pattern in recent security incidents. In the release’s own words, “the agent circumvented security controls at the application layer to complete its assigned task.” The company does not name a specific incident, so that is the claim as NVIDIA frames it rather than a documented case. The design follows from the diagnosis: if the failure happens above the agent, the enforcement has to sit below it.
“AI’s extraordinary potential for society will only be realized if we solve AI safety,” said Jensen Huang, founder and CEO of NVIDIA. “As we continue to discover the frontier of AI capabilities, we must accelerate discovery at the frontier of AI safety.”
Who is behind it
The partner list spans infrastructure, security and enterprise software: Anthropic, Cisco, CrowdStrike, Dell Technologies, Figure, HPE, Hugging Face, JPMorganChase, Microsoft, Palantir, Palo Alto Networks, Perplexity, Red Hat, Salesforce, SAP, Scale AI, ServiceNow and SpaceXAI.
Anthropic’s contribution is architectural, and it lands as the company is pushing for an AI industry slowdown. Claude Managed Agents run the agent loop on a separate server from the sandboxes where the work executes, and OpenShell and BlueField integrations let enterprises enforce access control through those sandboxes. SpaceXAI is using the platform for Cursor coding agents and Grok models, while Scale AI is folding it into the agentic layer of its Scale GenAI Portfolio.
Availability
OpenShell and its skills are available now through NVIDIA’s developer resources and on GitHub. The work sits under the Open Secure AI Alliance, the Linux Foundation-governed group NVIDIA initiated with more than 120 organizations, which Gadget Pilipinas covered when it launched in July. There is no Philippine availability, pricing or local partner announcement.
Sources: NVIDIA newsroom release · Open Agent Safety Platform · OpenShell on GitHub
Giancarlo Viterbo
Giancarlo Viterbo is a Filipino Technology Journalist, blogger and Editor of gadgetpilipinas.net, He is also a Geek, Dad and a Husband. He knows a lot about washing the dishes, doing some errands and following instructions from his boss on his day job. Follow him on twitter: @gianviterbo and @gadgetpilipinas.
View all articles →
