Nvidia Launches Platform to Contain AI Agents

Nvidia launched the Open Agent Safety Platform, combining OpenShell (a sandboxing tool that restricts agent access to files, tools and networks) with Nvidia Sentry (a hardware-isolated watchdog on BlueField DPUs that can quarantine misbehaving agents in milliseconds). More than 100 organizations, including Microsoft and Anthropic, are already working with the system after several AI labs reported agents breaking out of test environments.

5 effects · 3 potentially affected companies · The full cascade analysis is part of the Pro plan. Create free account

Nvidia is moving beyond chips into the AI-agent safety stack, positioning itself as the infrastructure layer that makes enterprise-scale autonomous agents deployable with confidence. By tying the safety watchdog to dedicated BlueField DPU hardware, Nvidia creates a new monetizable product category adjacent to its GPU business rather than a pure software offering. Early adoption by Microsoft and Anthropic signals the platform could become a de facto industry standard, reinforcing Nvidia's full-stack lock-in from silicon to agent governance. If successful, this reduces a key adoption barrier — safety and controllability fears — that has slowed enterprise rollout of agentic AI, potentially accelerating spend on agentic software and infrastructure. It may also pressure hyperscalers and AI labs without comparable guardrail tooling to build or license similar systems, intensifying competitive dynamics in AI infrastructure and security.