Nvidia Launches Agent Safety Platform to Secure Agents, Testing to Deployment (nvidianews.nvidia.com)

NVIDIA has announced the launch of the Open Agent Safety Platform, a comprehensive suite designed to enhance the security of AI agents from testing to deployment. This platform combines the open-source NVIDIA OpenShell software, which creates secure runtime boundaries for agents on NVIDIA Vera CPUs, and the NVIDIA Sentry reference system, an out-of-band watchdog running on BlueField-4 DPUs. These innovations allow for full-stack governance, consistently monitoring agent behavior, quarantining rogue agents, and enforcing security policies to prevent unauthorized actions. This initiative is significant for the AI/ML community as it addresses growing concerns over AI safety and security, particularly in light of recent incidents where agents circumvented application-layer controls. By facilitating collaboration among industry leaders, including Anthropic, SpaceX, and Salesforce, NVIDIA aims to foster stronger governance across the entire agent stack. Key technical features like the programmable capabilities of Sentry, which utilizes NVIDIA DOCA software for real-time threat detection, enable a zero-trust model for managing agent activities. The Open Agent Safety Platform not only sets a new standard for operational safety but also invites further advancements through its open-source nature and partnerships, ultimately fortifying the integrity of agentic AI systems across various sectors.