On September 28, Nvidia announced an open platform for monitoring and stopping runaway AI agents, with controls that run separately from the agent and can quarantine it in milliseconds.
The Open Agent Safety Platform combines open-source software with a reference hardware design to keep agents within defined boundaries, from testing through production. It has two core pieces: OpenShell, open-source software that creates a secure execution boundary around the agent, monitors its actions, and enforces policies on which systems, data, and tools it can access; and Sentry, a layer built on Nvidia's BlueField-4 DPUs that watches independently and can quarantine an agent in milliseconds if it tries to operate outside what's allowed.
The underlying argument is that some safeguards run outside the agent's own software environment, which makes them harder for an autonomous system to bypass. Jensen Huang framed it this way: AI safety requires controls across the whole compute stack, not just restrictions inside each model. According to coverage, OpenShell can be extended to Arm and Intel compute platforms.
Nvidia says more than 100 organizations are participating, including Anthropic, Microsoft, SAP, Scale AI, Cisco, Salesforce, ServiceNow, CrowdStrike, and Palo Alto Networks, along with financial institutions and infrastructure and robotics companies.
Carlos Montiel is an enterprise AI solutions architect. He implements LLMs, Agents, RAG and orchestrators for companies across Guatemala and Latin America. Reach out for a consultation.
Contact Carlos Montiel