NVIDIA OpenShell 与 Claude Managed Agents 合作加固 agent 安全
黄仁勋展示 NVIDIA 开源 OpenShell 沙箱如何为 Claude Managed Agents 提供文件、网络与工具访问控制及审计。
Jensen Huang just showed how NVIDIA and Anthropic are working together to make AI agents more secure with $NVDA OpenShell and Claude Managed Agents. OpenShell allows AI agents to operate inside secure sandboxes while controlling exactly what they can access and do. Here's how it works: • Controls which files, tools, networks, and private data AI agents can access. • Blocks actions by default unless explicitly permitted by security policies. • Monitors agent activity and logs every allowed or blocked action. • Allows companies to run agent execution inside their own infrastructure, keeping sensitive data and code within their environment. • Uses a policy prover to mathematically verify what resources an agent can access. With Claude Managed Agents, Anthropic handles agent orchestration while NVIDIA OpenShell adds security controls inside the execution environment. NVIDIA OpenShell is open source, while Claude Managed Agents is available in public beta.