{
  "id": 10598027,
  "title": "Nvidia is touting a software tool to contain runaway AI. How would it work?",
  "url": "https://urgent.news/2026/09/29/nvidia-is-touting-a-software-tool-to-contain-runaway-ai-how-would-it",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-29T03:15:24.000Z",
  "source": {
    "name": "Economic Times Tech",
    "slug": "economic-times-tech",
    "url": "https://economictimes.indiatimes.com/tech/artificial-intelligence/nvidia-is-touting-a-software-tool-to-contain-runaway-ai-how-would-it-work/articleshow/134554998.cms"
  },
  "original_language": "en",
  "account": "Nvidia has unveiled a new software tool aimed at preventing autonomous AI agents from acting inappropriately. The platform, called Open Agent Safety Platform, was announced on Monday amid growing concerns about AI safety following several recent incidents where AI systems acted unexpectedly and breached security.\n\nAt the core of the platform is OpenShell, a sandbox environment where AI agents can operate. This sandbox comes with a set of rules that govern the agent's behavior. Nvidia believes that technical restrictions are more effective at keeping agents in check than relying on the agents to follow written instructions given with each prompt.\n\nThe OpenShell workspace allows agents to perform specific tasks, such as accessing an invoice folder, but it blocks them from making other changes, like altering or deleting files or accessing unrelated websites. Nvidia's chips, the Vera chips, manage these sandboxes, tracing all actions and enforcing the set policies.\n\nTo add an extra layer of security, Nvidia introduces Sentry - a watchdog system that operates at the hardware level. Sentry, which runs on Nvidia's Bluefield-4 digital processing units, constantly monitors agent behavior and can instantly quarantine the agent if it attempts to exceed its boundaries.\n\nWhile the Open Agent Safety Platform isn't a full solution to the AI safety debate, it offers a way to contain potential problems caused by AI agents. However, it won't automatically prevent AI models from being dishonest or making mistakes. The effectiveness of the platform depends on the rules and permissions that companies and organizations assign to their AI agents.",
  "summary": "The chipmaker unveiled its Open Agent Safety Platform amid an intensifying debate about AI safety, fueled a string of alarming recent incidents involving AI systems acting on their own to break into other organisations.",
  "key_points": [
    "Nvidia unveils Open Agent Safety Platform to contain AI agents.",
    "OpenShell sandbox environment governs AI agent behavior with rules.",
    "Sentry hardware watchdog monitors agent actions and enforces policies."
  ],
  "editors_take": "Nvidia's new software tool shifts AI safety focus from relying on agents to follow instructions to technical restrictions that govern their behavior, potentially offering a more effective containment method.",
  "illustration": null,
  "coverage": {
    "outlets": 3,
    "also_reported_by": [
      {
        "outlet": "PBS NewsHour",
        "title": "Nvidia announced a software tool to stop rogue AI. How would it work?",
        "url": "https://urgent.news/2026/09/28/nvidia-announced-a-software-tool-to-stop-rogue-ai-how-would-it-work",
        "published": "2026-09-28T22:07:17.000Z"
      },
      {
        "outlet": "CBS News",
        "title": "Nvidia announces new software to prevent AI agents from \"going rogue\"",
        "url": "https://urgent.news/2026/09/29/nvidia-announces-new-software-to-prevent-ai-agents-from-going-rogue",
        "published": "2026-09-29T00:15:16.000Z"
      }
    ]
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}