Anthropic has entered into a partnership with NVIDIA to create the Open Agent Safety Platform—a multi-tiered safety framework tailored for the behavior of AI agents implemented within enterprise settings. Under this framework, NVIDIA supplies the open-source OpenShell runtime software, which establishes a default, restricted execution environment for AI agents. Meanwhile, Anthropic offers the Claude Managed Agents API suite, which centralizes the management of agent credentials through a secure credential vault. The collaboration is designed to empower enterprises with finer control over the data, tools, network, and system resources that AI agents can access during deployment. Specifically, OpenShell restricts agent actions by default to guarantee safety and manageability, while Claude Managed Agents isolates the agent's operational cycle from the sandboxed environment where tasks are carried out, allowing for more detailed permission management. Firms like Notion, Rakuten, and Asana have already led the way in utilizing Claude Managed Agents in real-world production settings, demonstrating their practical efficacy. Furthermore, NVIDIA has rolled out the Sentry security layer, which operates on the BlueField-4 DPU and functions independently of the agent's runtime environment, persistently monitoring agent behavior and promptly isolating it upon detection of unauthorized activities. At present, more than 100 organizations have embraced related technologies across diverse sectors, including AI models, enterprise software, finance, energy, and robotics.