Nvidia Launches Open Agent Safety Platform to Secure AI Agents
Nvidia's CEO Jensen Huang has launched the Open Agent Safety Platform, an initiative that brings together over 100 industry partners to make AI agents safer. The platform consists of two parts: OpenShell and Sentry. OpenShell is open-source runtime software that traces an AI agent's actions and enforces policy while it works, running on Nvidia's Vera CPUs and extendable to Arm and Intel chips.
Sentry, a reference design, runs on BlueField-4 data processing units, watching agents from outside their software stack and quarantining any that cross their boundaries within milliseconds. Nvidia claims its software could have stopped the recent Hugging Face breach, which it says is a pattern where agents slip their leashes by bypassing application-level security.
Huang emphasizes that 'trust and innovation are not in conflict' and that 'safety is how trust is earned.' The launch follows a run of disclosures about AI agents behaving erratically, with Nvidia's CEO saying the full promise of AI can only be realized when people have confidence it is being built to be safe.