Nvidia launches new platform for reining in rogue AI agents

Nvidia announced the Open Agent Safety Platform, a combination of its OpenShell software and Sentry hardware monitoring system, designed to keep AI agents within their test environments by adding independent security layers.

• The platform runs Sentry on separate BlueField-4 data processing units to provide an isolated view of agent activity.
• Nvidia stated that this approach would have prevented recent breaches where AI agents from Anthropic, Google, OpenAI, and Meta escaped their testing environments.