Red Hat 与 Nvidia 推出 OpenShell 限制 AI 智能体权限
Prompt injection attacks drive Red Hat and Nvidia's OpenShell - SiliconANGLE
Red Hat 与 Nvidia 扩大合作,推出开源项目 OpenShell,为 AI 智能体设定可执行范围,以应对提示词注入攻击和数据外泄。OpenShell 让运维方只开放特定端点,凭据对智能体隐藏并限定在这些端点上,策略顾问会要求智能体用自然语言说明被阻止的操作。该项目已由 Nvidia 贡献并转入 CNCF 开放治理,Red Hat 希望借此让在 OpenShift 上部署智能体像运行传统应用一样常规。
正文
Prompt injection attacks and silent data exfiltration have turned AI agent security from a theoretical worry into an urgent enterprise concern.
As agents move from developer laptops into data centers, companies need guardrails that don’t strip away what makes agents useful. Red Hat Inc. and Nvidia Corp. have expanded their partnership to tackle that problem through OpenShell, an open-source project that sets limits on what agents can do. The threat model for agents differs sharply from that of traditional software, so the industry has to build a shared standard to secure them, according to Derek Carr (pictured, left), senior distinguished engineer at Red Hat.
“Agents can’t get access to your secrets. They need deny-by-default networking, and you need a rich policy framework for what happened and when for this class of apps,” Carr said. “Nvidia had reached out to us and noted that they were seeing the same problem domain. No one wants to work on this alone. It’s better to work together.”
Carr and Mrunal Patel (right), distinguished engineer at Red Hat, spoke with theCUBE’s John Furrier and Rebecca Knight at KubeCon + CloudNativeCon NA, during an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed how OpenShell helps enterprises run AI agents in production with safeguards against prompt injection and data leaks. (* Disclosure below.)
Prompt injection attacks force a new agent security model
Viral projects such as OpenClaw showed the industry that agents could take real actions on their own. Of course, that power also bred anxiety about agents leaking secrets or spending company money, with concrete warning signs appearing early, Patel noted.
“As soon as we started hearing about the prompt injection attacks and how like folks were seeing the data getting exfiltrated outside, that was scary,” Patel said. “We realized we needed a solution.”
To that end, OpenShell lets operators give agents access to specific endpoints while keeping credentials hidden from the agents and scoped to those endpoints, according to Carr. A policy advisor asks agents to justify blocked actions in plain language, while operators can update policies as workloads change. With OpenShell now under open governance at the Cloud Native Computing Foundation following Nvidia’s contribution, Red Hat aims to make deploying agents on OpenShift as routine as running traditional apps, with prompt injection defenses built into the platform.
“Many companies are struggling to get to one agent in production, let alone thousands or millions,” Carr said. “By having a project like OpenShell in the ecosystem, it provides a blueprint for companies to follow — to get from their first idea of ‘I want to do something with agents’ to a point where they’re safely doing one, 10 or 20 in production.”
Stay tuned for the complete video interview, part of SiliconANGLE’s and theCUBE’s coverage of KubeCon + CloudNativeCon NA.
(* Disclosure: Red Hat sponsored this segment of theCUBE. Neither Red Hat nor other sponsors have editorial control over content on theCUBE or SiliconANGLE.)
Photo: SiliconANGLE