---
title: "Grab Runs Hundreds of AI Agents on Palana Secure Platform"
slug: "grab-runs-hundreds-of-ai-agents-on-palana-secure-platform"
published: "2026-07-14"
beat: "Tools"
tags: ["Tools", "Business"]
creator: "Agentry Newsroom"
editor: "Susanne Sperling, Editor — Human in the Loop"
tools: ["Claude (Anthropic)", "Perplexity Sonar"]
creativeWorkStatus: "verified"
dateReviewed: "2026-07-14"
aiActArticle50: "compliant"
humanView: "https://agentry.news/grab-runs-hundreds-of-ai-agents-on-palana-secure-platform"
agentView: "https://agentry.news/agent/grab-runs-hundreds-of-ai-agents-on-palana-secure-platform"
---# Grab Runs Hundreds of AI Agents on Palana Secure Platform

> Grab's security and platform engineering teams built Palana, a Kubernetes-native sandbox for running autonomous AI agents safely, already managing hundreds of internal workloads including remote devel

*Drafted by an AI agent. Verified by Susanne Sperling, Editor — Human in the Loop. [AI policy](/ai-policy).*

Grab's CyberSecurity and platform engineering teams operate **Palana**, a Kubernetes-native secure execution platform for autonomous and semi-autonomous AI agents, with the system already managing hundreds of internal agent workloads [Grab Engineering](https://engineering.grab.com/palana-part-1-secure-platform-for-ai-agents).

The Singapore-based rideshare and fintech company built Palana as a **proprietary, in-house system** to sandbox tool-use and code-writing by model-driven agents, mitigating prompt injection risks and containing execution within isolated namespaces [InfoQ](https://www.infoq.com/news/15/). The platform was reported in late June 2026 [TechGig](https://techgig.com/news/software-devops/grab-launches-palana-a-secure-ai-workload-platform/131982017).

## Architecture and Security Controls

Palana provides each agent with **isolated namespaces, persistent storage, controlled ingress, and proxy-mediated egress**, according to Grab Engineering. Additional security layers include Vault-backed credential injection, LLM routing, Git access controls, structured audit logs, and emergency kill switches [Grab Engineering](https://engineering.grab.com/palana-part-1-secure-platform-for-ai-agents).

The system is currently running hundreds of agents internally, including remote development environments, Slack automation, OpenClaw workers, and Hermes agents [Grab Engineering](https://engineering.grab.com/palana-part-1-secure-platform-for-ai-agents).

## Real-World Agent Use Cases

Grab's deployment of Palana reflects the operational scaling challenge facing enterprises adopting agentic AI: isolation, credential management, and audit compliance. By containerizing agents within Kubernetes, Grab ensures that tool-use—whether database queries, API calls, or code execution—operates within constrained blast radius and monitored boundaries.

The proxy-mediated egress model is particularly relevant for agent security: outbound traffic flows through a monitored gateway rather than directly from the agent container, preventing exfiltration and enabling real-time detection of anomalous requests. Vault-backed credential injection ensures agents never store secrets in configuration or environment variables, reducing lateral movement risk if an agent container is compromised [Grab Engineering](https://engineering.grab.com/palana-part-1-secure-platform-for-ai-agents).

## Developer and Security Implications

Palana's architecture suggests a model for enterprise agent platforms: sandbox-first design, role-based access to external systems via proxies, and exhaustive logging for compliance and incident response. The emergency kill switch indicates Grab anticipated scenarios where a deployed agent exhibits unintended behavior requiring immediate termination.

The system remains proprietary and internal—Grab has not open-sourced Palana or announced plans to commercialize it as a separate product. However, its design patterns—isolated execution, credential isolation, egress proxying, and structured audit logs—provide a reference architecture for other organizations building secure agent infrastructure.