NVIDIA and Anthropic Launch AI Agent Safety Platform

NVIDIA and Anthropic have announced the Open Agent Safety Platform, combining NVIDIA's open source OpenShell runtime with Anthropic's Claude Managed Agents to give companies layered control over AI agents. Managed Agents stores credentials in a vault so agents never access them directly, and provides audit trails and access control integration. OpenShell governs agent behavior by blocking all actions unless explicitly permitted, logging every decision, and using mathematical proof to verify what an agent can reach. Customers including Notion, Rakuten, and Asana are already using Managed Agents. OpenShell is available on GitHub under the Apache 2.0 license.