OpenClaw Releases Free Enterprise Platform for Secure AI Agent Deployment

OpenClaw has introduced OpenClaw Enterprise (OCE), a free, open-source platform aimed at enabling organizations to run persistent AI agents with improved security and centralized control. Announced on September 29, 2026, OCE addresses the growing need for AI agents to perform ongoing tasks without granting them unrestricted access to sensitive enterprise systems. While still under active development with a planned 1.0 release later this year, OpenClaw recommends OCE for internal pilot workloads rather than immediate broad production deployment.

The core of OCE is its enterprise control plane, a central layer that facilitates the deployment of AI agents, the establishment of operational rules, and the comprehensive tracking of their activities. The platform supports multi-tenancy, allowing for distinct security boundaries between trusted services and potentially untrusted agent workloads. OpenClaw emphasizes that OCE's security architecture integrates multiple layers of defense, including sandboxing, fine-grained permission controls, and reviews conducted by large language models themselves. These measures are designed to strictly limit agent access and actions while preserving their utility.

Governance and audit trails are maintained throughout the agent lifecycle, providing operators with the ability to trace actions and review decisions made by the AI. This comprehensive approach tackles a significant concern highlighted by OpenClaw: the lack of shared security, safety, and governance standards for AI agents, which often leads organizations to block their adoption. Agents capable of accessing internal data, executing code, or utilizing plugins require robust controls that go beyond simple instructions for safe behavior.

The practical necessity for such controls is underscored by past incidents. Cybersecurity News has previously reported on data leaks facilitated by indirect prompt injection, where hidden malicious instructions within external content could trigger the disclosure of sensitive information. Separate incidents involving log poisoning demonstrated how attacker-controlled log entries could manipulate an agent's context during troubleshooting. These past issues, while not directly related to newly disclosed OCE vulnerabilities, inform the security design of the new platform.

The OCE project originated at OpenAI before being donated to the independent OpenClaw Foundation. Development is now a collaborative effort involving Red Hat and NVIDIA. The platform is released under the permissive MIT license, ensuring it remains free for organizational use. Vendor neutrality is a key design principle, allowing organizations to substitute components like the underlying AI model, sandbox environment, and agent harness with their own preferred or existing solutions, thereby avoiding vendor lock-in.

Organizations can self-host OCE by utilizing its public GitHub repository and accompanying documentation. While Kubernetes is supported for internal deployments, the current repository notes that the default Compose preview runs the control plane but does not deploy agents; a separate Kubernetes profile is used for local agent walkthroughs. It's important to note that while the software is free, organizations will still incur infrastructure and model-serving costs.

OpenAI and Red Hat are reportedly piloting OCE internally. For instance, an OpenAI staff member, RJ Marsan, described an internal enterprise agent named Androidclaw that assists in investigating build failures, locating incident records, and preparing fixes with supporting evidence. These use cases highlight the critical importance of access control, as an agent connected to code repositories, logs, and collaboration tools can operate across multiple sensitive systems.

For security teams, OCE offers the immediate benefit of an inspectable and testable open-source platform. However, given its early stage of development, organizations are advised to thoroughly verify isolation mechanisms, permission settings, audit coverage, and plugin behavior before granting agents access to critical workloads. The foundation plans to release a security reference architecture in the near future.