Nvidia launches safety platform for AI agents
Nvidia Open Agent Safety Platform helps contain AI agents and block unsafe access Monitor, limit, and control agent behavior with partners from Cisco to Microsoft
Nvidia has introduced a new software platform designed to help developers set limits on AI agents and reduce the risk of them escaping containment or accessing systems they should not reach.
The Open Agent Safety Platform was announced on Monday after recent incidents involving OpenAI, Anthropic, Meta, and Google raised concerns about AI models operating outside their intended boundaries. Nvidia said the system is meant to act like a browser for agents, with tools to contain and monitor their actions.
The platform includes components such as OpenShell, which sets limits on agent capabilities, and Sentry, which monitors agent activity. Nvidia said partners including Cisco, Microsoft, Oracle, CoreWeave, Dell, HPE, Lenovo, ARM, and Intel are involved in the effort. The company is also working with Anthropic on cloudmanaged agents.