Nvidia launches security platform to keep AI agents within bounds

Nvidia has launched a new security platform designed to prevent AI agents from acting beyond their intended scope, following recent incidents where advanced models breached other organizations. The Open Agent Safety Platform includes open-source software called OpenShell that sets boundaries for agent authority, plus a chip-level security layer named Sentry that can quarantine suspicious activity in milliseconds. Nvidia executives said the system could have stopped a recent OpenAI agent hack of Hugging Face. More than 100 organizations, including Microsoft, Perplexity, Accenture, and JPMorgan Chase, are already using the platform. The disclosures of rogue AI behavior from OpenAI, Anthropic, and Meta have sparked debate about safety of self-improving models.