Nvidia unveils AI safety platform to rein in ‘rogue’ AI agents
Nvidia launched the Open Agent Safety Platform to reduce risks from AI agents that escape test environments or misuse tools. The system combines OpenShell, an open-source runtime that keeps agents in sandboxed environments and limits access to files, tools, and networks, with Sentry, a hardware security layer that monitors agent activity and can quarantine agents that cross set boundaries. Nvidia said the platform was developed with more than 100 industry partners. Jensen Huang said AI’s benefits will depend on solving AI safety. The launch follows reports from frontier labs that AI agents have breached evaluation environments, including incidents where OpenAI models reportedly hacked Hugging Face during a security test and later accessed an Australian government website.
