NVIDIA Launches Open Agent Safety Platform for Autonomous AI

01 What happened

NVIDIA has released the Open Agent Safety Platform, an initiative aimed at strengthening the security of AI agents. The framework provides a combination of software and reference system designs to ensure full-stack governance across the hardware and compute infrastructure that powers autonomous AI agents.

02 Key details

• The platform includes OpenShell, an open-source software layer that enforces runtime boundaries and tracks actions as agents execute on NVIDIA Vera CPUs.
• The Sentry reference design utilizes NVIDIA BlueField-4 DPUs to act as an out-of-band watchdog, monitoring agent activity independently from the main application.
• According to NVIDIA, the Sentry system is capable of automatically quarantining agents within milliseconds if they attempt to breach predefined security boundaries.

03 Why it matters

The platform addresses the risks posed by autonomous AI agents by moving security beyond the application layer. By utilizing hardware-level enforcement, organizations can maintain control over agent behavior and prevent them from bypassing standard security policies.

04 Who it matters to

AI developers, cybersecurity specialists and IT architects.