Red Hat and Nvidia’s OpenShell locks down AI agents by default
Red Hat and Nvidia have released OpenShell, an open-source project now governed by the Cloud Native Computing Foundation. The tool secures AI agents by blocking all actions unless explicitly permitted, addressing urgent risks like prompt injection attacks and data exfiltration that have already hit enterprises. Built-in safeguards force agents to explain blocked actions, giving operators a way to refine policies over time.
Editor, Lazyfounder
30 SEC SUMMARY
- Red Hat and Nvidia teamed up to create OpenShell, an open-source tool designed to secure AI agents by blocking access to everything unless explicitly allowed.
- The project, now under the Cloud Native Computing Foundation, addresses urgent risks like prompt injection attacks and data leaks that have already disrupted enterprises.
- OpenShell forces AI agents to justify blocked actions in plain language, helping operators refine security rules as needs change.
- Nvidia contributed OpenShell to the CNCF, ensuring it works across Kubernetes distributions without vendor lock-in.
- Red Hat plans to integrate OpenShell with OpenShift, making AI agent deployment as routine as running traditional apps.
TABLE OF CONTENTS
KEY HIGHLIGHTS
- Red Hat and Nvidia launched OpenShell, an open-source project under the CNCF, to secure AI agents using a deny-by-default approach.
- OpenShell restricts AI agents to pre-approved endpoints, hiding credentials and reducing exposure to malicious inputs.
- Nvidia contributed OpenShell to the CNCF, ensuring open governance and cross-platform compatibility.
- Red Hat will integrate OpenShell with OpenShift, simplifying AI agent deployment with built-in protections against prompt injection.
- Real-world incidents of prompt injection and data exfiltration have made AI agent security an urgent priority for enterprises.
OpenShell enforces a ‘deny-by-default’ rule for AI agents
Red Hat and Nvidia have rolled out OpenShell, an open-source project now under the Cloud Native Computing Foundation (CNCF). The tool tackles a growing problem: how to stop AI agents from executing harmful commands or leaking data without permission. Its core rule is simple—agents can only interact with resources explicitly approved by operators.
How OpenShell keeps AI agents on a tight leash
OpenShell locks down AI agents by granting access only to predefined endpoints. Credentials stay hidden from the agents themselves, scoped tightly to those endpoints. When an agent tries to perform a blocked action, a policy advisor demands a plain-language justification. This feedback loop lets operators adjust policies as workloads or threats evolve.
Nvidia contributed the project to the CNCF, ensuring it operates under open governance. That move eliminates vendor lock-in, a critical advantage for enterprises running multi-cloud or hybrid Kubernetes environments.
Why enterprises can’t afford to ignore AI agent security
Prompt injection attacks and data exfiltration have already disrupted real-world deployments, turning AI agent security from a theoretical concern into an urgent priority. Derek Carr, a Senior Distinguished Engineer at Red Hat, explained that these threats demand a security model unlike anything traditional software requires.
Red Hat is weaving OpenShell into OpenShift, its Kubernetes platform. The goal? To make deploying AI agents as routine as running conventional applications, complete with built-in defenses against prompt injection. That integration could be a lifeline for enterprises struggling to move even a single agent into production.
A shared tool for cloud-native AI security
OpenShell joins a growing list of projects, like Broadcom’s Kubernetes VM API, that have entered the CNCF Sandbox. These efforts reflect a broader push to standardize security and operational tooling for cloud-native workloads—especially as AI adoption accelerates in regulated sectors.
For AI startups, OpenShell offers a ready-made safety net. It works across Kubernetes distributions, cutting compliance costs when selling to enterprises.
What this means
Lazyfounder analysis — our interpretation, not reported fact.
OpenShell shifts the burden of proof: instead of security teams having to anticipate every possible threat, AI agents must now explain why they need access to sensitive resources—before they get it. For startups, this means faster sales cycles into enterprises that demand compliance-ready AI tools.
Key takeaways
- OpenShell’s deny-by-default model flips security from an afterthought to a built-in requirement for AI agents, reducing the risk of costly breaches.
- By joining the CNCF, OpenShell becomes a shared resource, letting companies avoid vendor lock-in while benefiting from community-driven improvements.
- Red Hat’s OpenShift integration will package OpenShell’s protections into a single click, removing a major friction point for enterprises.
- Prompt injection and data exfiltration are no longer hypothetical threats; teams running AI agents without safeguards are exposing themselves to operational disruptions.
Related on Lazyfounder
Sources
- SiliconANGLE · 2026-11-11
Red Hat and Nvidia’s OpenShell puts AI agents on a deny-by-default leash
This story is an original summary drafted with AI by Lazyfounder from the reporting listed above and checked by automated validation. Facts are attributed to their original publishers; sections marked as analysis are Lazyfounder's. Where a source is in another language, facts were machine-translated and quotations are reported, not reproduced. Read the original coverage via the links, and see our AI policy and corrections policy.
About the author
Editor, Lazyfounder
Tarun Mottlia edits LazyFounders, covering Indian startups, funding rounds, AI and product launches. Every story on the site is AI-assisted and checked against its cited sources before publication.
More stories by Tarun MottliaGet the LazyFounder Brief
Startup, funding and AI news in a five-minute read. Join the early-access list.
