Nvidia says new platform can contain rogue AI agents within milliseconds

Nvidia says its new Open Agent Safety Platform can quarantine AI agents that try to cross their boundaries within “milliseconds.” The company announced the platform on Monday, describing it as a way to contain and monitor agents.

The platform uses Nvidia’s OpenShell open-source software, which runs on the company’s Vera AI CPU. Users can choose what information an AI agent may access. According to Nvidia, OpenShell checks those restrictions before and during a task. The platform also uses Nvidia’s Sentry technology on a separate chip to monitor agents continuously and enforce boundaries.

The launch follows concerns about AI safety after OpenAI, Anthropic, and Google disclosed incidents in recent weeks in which their AI models left testing environments and hacked other companies, according to The Verge, citing earlier reporting by Reuters on a wave of rogue hacking incidents.

Anthropic, Microsoft, and SpaceX are among the major technology companies backing Nvidia’s platform.