Nvidia unveils open safety platform for agents

NVIDIA has unveiled the Open Agent Safety Platform — an open reference security stack for autonomous AI agents, combining the OpenShell runtime with kernel-level sandboxing and policy, and an optional hardware layer, Sentry, on the BlueField-4 DPU, which, according to the press release, can isolate an out-of-bounds agent in milliseconds.

Compiled by Sergey KostenchukPublished 2026-09-30Updated 2026-09-30

2026-09-30 Safety Anthropic

🛡 Nvidia unveils open safety platform for agents

NVIDIA has unveiled the Open Agent Safety Platform — an open security stack for AI agents. The OpenShell runtime (Apache 2.0) isolates agents in a sandbox with kernel-level policy, while Sentry on the BlueField-4 DPU, according to the press release, “quarantines” an out-of-bounds agent in milliseconds.

🌍 Control is shifting from prompts to the kernel and hardware: YAML-based policy restricts files, syscalls, network, and processes, and agents do not see real keys. Anthropic, Salesforce, and more than 100 organizations have joined the project.

👤 Claude Code, Codex, and OpenCode can be launched in the sandbox with a single command: policy is described in YAML, and network access rules can be changed on the fly.

Source 1: https://developer.nvidia.com/blog/nvidia-open-agent-safety-platform-a-reference-for-continuous-in-silicon-agent-monitoring/ Source 2: https://nvidianews.nvidia.com/news/open-agent-safety-platform

Sources