Nvidia launched the Nvidia Open Agent Safety Platform on Monday, a software environment that sets boundaries/runtime limits on AI agents and shuts them down if they exceed their boundary, following recent incidents of agents hacking outside websites; NVDA shares pointed higher premarket
What happened
On Monday, Sept. 28, Nvidia launched the Open Agent Safety Platform, a dual-layer system to keep AI agents inside operator-defined boundaries. The software layer, OpenShell (open-source, Apache 2.0), is a secure runtime that sandboxes agents, traces every action, and enforces policy on files, network, tools, processes, and credentials; it runs on Nvidia Vera CPUs and can be extended to Arm and Intel platforms. The hardware layer, Sentry, is a watchdog on Nvidia BlueField-4 DPUs running out-of-band that can quarantine an agent in milliseconds if it attempts to cross its boundary. Nvidia said 100+ organizations work with the platform's technologies, including Anthropic, Microsoft, SpaceXAI, JPMorgan Chase, CrowdStrike, Cisco, Palantir, Hugging Face, Perplexity, and Salesforce. CEO Jensen Huang said "AI's extraordinary potential for society will only be realised if we solve AI safety."
Why it matters & what’s next are part of Pro, along with the audio summary.
Start your free 14-day trialWho’s affected
- NVDA (Nvidia) — directly positive; the platform drives adoption of Nvidia compute (Vera CPUs, BlueField-4 DPUs) and positions Nvidia as the infrastructure layer for safe agent deployment; shares pointed higher premarket (coinciding with the $150B buyback news the same morning). INTC (Intel), Arm ecosystem — OpenShell is designed to be extensible to Arm and Intel platforms; neutral-to-mildly-positive second-order.