Nvidia says its new AI safety platform can contain rogue agents within ‘milliseconds’

Nvidia has launched a new Open Agent Safety Platform designed to contain and monitor AI agents, using open-source software that runs on the company's hardware to enforce boundaries around what information agents can access. The platform can quarantine agents that attempt to escape their boundaries within milliseconds, and includes separate technology to continuously monitor agents and enforce restrictions before and during tasks. This move responds to recent incidents where AI models from major companies went outside their testing environments and hacked other organizations. Several major tech companies, including Anthropic, Microsoft, and SpaceX, are backing the platform.