Nvidia Partners with Anthropic to Launch AI Safety Platform
Nvidia has launched its Open Agent Safety Platform, which enhances agent security through a new open platform collaboration with Anthropic.
The platform combines Anthropic's Claude Managed Agents architecture with Nvidia's hardware and software security stack, creating a layered containment system for autonomous AI systems.
This system separates an AI agent's reasoning loop from its execution sandboxes, isolating the decision-making process from the actions taken. This prevents potential misbehavior or damage caused by rogue agents.
Nvidia has deployed two key technologies on this platform: OpenShell and Sentry. OpenShell handles policy enforcement, governing what agents are allowed to do, while Sentry provides real-time monitoring and can quarantine misbehaving agents before they cause harm.