NVIDIA launches OpenShell to control what an AI agent can access

Giving an AI agent access to your data and tools now comes with a new layer of control.

On September 28, NVIDIA announced Open Agent Safety Platform, an open platform to protect agents from testing all the way to deployment. It includes OpenShell, an open-source environment that sets which systems and data an agent can access, without having to rewrite the agent. It is already generally available and works with coding agents such as Codex and Claude Code. Companies such as Anthropic, Microsoft, Salesforce and SAP are joining the initiative.

For a small business that is starting to hand tasks over to agents, the risk is not only that they make mistakes, but that they access things they should not. At InfinyAI we believe that setting limits outside the agent itself will become more and more common.

How do you control what your agents do?