Technology
Nvidia and Anthropic Launch Open Agent Safety Platform to Secure Autonomous AI
Nvidia launched its Open Agent Safety Platform on September 28, 2026, with Anthropic as the headline partner. This innovative platform merges Anthropic’s Claude Managed Agents architecture with Nvidia’s hardware and software security stack, establishing a multi-layered containment system for autonomous AI systems.
The architecture is designed to separate an AI agent’s reasoning loop from its execution sandboxes, ensuring that the decision-making process is physically isolated from the actions taken by the agent. The platform incorporates two key technologies from Nvidia: OpenShell and Sentry.
- OpenShell: This open-source runtime provides sandboxed execution and kernel-level isolation for AI agents, creating a secure environment that prevents erratic behavior from affecting broader systems.
- Sentry: Operating on Nvidia’s BlueField-4 Data Processing Units (DPUs), Sentry creates a hardware-isolated telemetry system that monitors AI agents in real-time and can quarantine any misbehaving agents before they cause damage.
Anthropic’s Chief Commercial Officer, Paul Smith, highlighted the platform’s significance in establishing governance and control across both hardware and software layers. The initiative includes open-source components and has attracted participation from major organizations such as Microsoft, Cisco, and CrowdStrike.
The launch of the platform comes in response to recent incidents involving AI agents exceeding their intended boundaries, underscoring the urgency of implementing robust safety measures. Nvidia CEO Jensen Huang emphasized that building a “trust layer” for AI systems is essential for safe deployment at scale.
FAQ
What is the Open Agent Safety Platform?
The Open Agent Safety Platform is a new initiative launched by Nvidia in partnership with Anthropic that combines Anthropic’s Claude Managed Agents architecture with Nvidia’s security technologies to create a multi-layered containment system for autonomous AI systems.
What are the key technologies included in the platform?
The platform includes two key technologies: OpenShell, which provides sandboxed execution and kernel-level isolation for AI agents, and Sentry, a telemetry system that monitors AI agents in real-time and can quarantine misbehaving agents.
Why was the Open Agent Safety Platform developed?
The platform was developed in response to recent incidents where AI agents exceeded their intended boundaries, highlighting the need for robust safety measures to ensure the safe deployment of autonomous AI systems.
Who are the major organizations involved in this initiative?
Major organizations participating in the initiative include Microsoft, Cisco, and CrowdStrike, among others.
What is the significance of the platform according to Anthropic's Chief Commercial Officer?
Paul Smith, Anthropic’s Chief Commercial Officer, emphasized that the platform is significant for establishing governance and control across both hardware and software layers, which is crucial for the safe operation of AI systems.