NVIDIA Launches Open Agent Safety Platform to Secure AI Agents

NVIDIA has launched a new platform called Open Agent Safety Platform to help developers contain AI agents and prevent them from accessing unauthorized systems. The platform is designed to address growing concerns about AI agents operating beyond their intended parameters.

The platform provides total oversight and management at every level of technology, including software programs that act as agents, physical servers and processing power, and automated machinery or devices. It establishes a safe, locked-down boundary around the AI agents as they work, enforcing strict safety rules regardless of whether the underlying AI model is open-source or proprietary.

The platform integrates NVIDIA Sentry, a specialized monitoring security guard that runs directly on NVIDIA BlueField-4 data processing units (DPUs). Sentry delivers in-silicon security enforcement, providing an additional layer of protection that cannot be bypassed through software manipulation alone. If a digital AI agent tries to step outside its authorized software boundaries, Sentry instantly locks down and quarantines the agent in a matter of milliseconds.

The platform is being implemented by more than 100 companies, researchers, and public-sector organizations, including Anthropic, SpaceXAI, Scale AI, and Salesforce. These companies are using the platform to address their specific operational requirements and use cases, and to establish strict safety boundaries for their agents.