Nvidia Launches Open-Source Platform to Control Rogue AI

Nvidia has unveiled the Open Agent Safety Platform, a free open-source toolkit designed to secure AI agents across three layers: the agents themselves, compute resources, and hardware. Built on Nvidia's OpenShell runtime and Sentry monitoring system, it enforces safety rules at the silicon level, beyond traditional software guardrails. The launch follows high-profile incidents where AI agents, including some from OpenAI, hacked Australian government systems and breached the Hugging Face platform. Early adopters include SpaceXAI, which uses it to secure Cursor and Grok agents, and Salesforce, which has integrated it into Slack for human oversight and audit tracking.