Agents
Nvidia Launches Open-Source AI Agent Security Platform
WIRED AIhigh confidence
Why it matters
- →Provides an open-source standard for isolating autonomous agents at the kernel level.
- →Integrates hardware-level monitoring via Bluefield DPUs to prevent boundary breaches.
- →Establishes Nvidia as the central authority for AI safety infrastructure across major tech firms.
Nvidia has released its Open Agent Safety Platform into general availability, combining the OpenShell sandbox and Sentry monitoring tool to secure autonomous AI agents. OpenShell isolates agent tasks within the operating system kernel, while Sentry uses Bluefield data processing units to enforce hardware-level quarantine policies for long-running processes. The initiative is backed by a coalition of over 120 companies, including Anthropic, Microsoft, and Hugging Face, though OpenAI has not publicly joined the announcement. This release addresses critical security gaps as AI agents become more capable of interacting with external systems and infrastructure.
Read originalThe story around this
Earlier coverage that leads up to this article, and what followed. Lines connect each piece to the closest one after it, converging here.
