Skip to main content
These terms (the “Sandbox Terms”) supplement and are hereby incorporated into the CoreWeave Terms of Service (“TOS”) and govern Customer’s use of, or access to, CoreWeave Sandbox (the “Sandbox Service”) which is considered “CoreWeave Services” as defined under the TOS. In event of a conflict or inconsistency with the TOS, these Sandbox Terms control solely with respect to the Sandbox Service. Capitalized terms not defined here have the meanings given in the TOS.

​
Sandbox Service

  1. Sandbox Service. The Sandbox Service enables Customer to create and manage isolated compute environments (each, a “Sandbox”), which Customer may use to run Agents, models, software, code, scripts, processes, and other workloads provided, selected, or controlled by Customer or a third party (collectively, “Customer Workloads”). Customer is responsible for all activity undertaken through or initiated by a Customer Workload in connection with the Sandbox Service as if undertaken directly by Customer. “Agent” means any artificial intelligence model, system, or workflow capable of taking actions or performing tasks on behalf of Customer autonomously or with human supervision.
  2. Customer Workloads. Customer is solely responsible for selecting, configuring, instructing, and operating Customer Workloads, including all credentials, secrets, permissions, network destinations, data, and external systems it makes available to Customer Workloads. Customer will not, and will not permit any Customer Workload to attempt to escape a Sandbox; defeat isolation or security controls; obtain unauthorized access to CoreWeave’s or its affiliates’, or any third party’s infrastructure or resources; or interfere with the operation or security of the Sandbox Service or CoreWeave Services. Customer will use commercially reasonable measures to limit Customer Workloads’ access to credentials, data, networks, and systems to what is necessary for the intended task, and will promptly revoke or rotate any credential used with the Sandbox Service that it knows or reasonably suspects has been compromised. Customer will obtain CoreWeave’s prior written consent before conducting any security or malware research on or using the Sandbox Service, including testing its isolation boundary or CoreWeave’s or its service providers’ systems. Customer will promptly notify CoreWeave if it becomes aware of any actual or reasonably suspected escape by a Customer Workload from a Sandbox, attempted defeat of the Sandbox Service’s isolation controls, or material vulnerability in the Sandbox Service.
  3. Sandbox Operations. Customer is solely responsible for: (i) determining whether the Sandbox Service and its selected configurations are appropriate for its intended use case; (ii) evaluating results before relying on them or using them outside the Sandbox; (iii) configuring, monitoring, stopping, and decommissioning its Sandboxes and workloads, including all usage initiated or continued automatically through a Customer Workload; and (iv) exporting content it wishes to retain before a Sandbox becomes unavailable, subject to the functionality described in the Documentation.