Security.

Our approach

Reverie is being designed around least-privilege access, workload isolation, auditable activity, and explicit customer control over research inputs and outputs. This page describes our direction, not a certification or a contractual commitment.

Platform safeguards

• Role-appropriate access to projects and administrative actions
• Isolated execution environments for model-training workloads
• Logs and experiment lineage to support review and investigation
• Managed handling of model artifacts, checkpoints, and reports
• Dependency and infrastructure maintenance as the platform evolves

Project boundaries

Each project begins with an explicit objective, action space, dataset, constraints, compute budget, and stopping conditions. Teams remain responsible for ensuring they have the rights to provide data and for reviewing outputs before operational use.

Due diligence

Security requirements vary by organization and domain. We welcome detailed discussions about architecture, deployment needs, data handling, access controls, and review procedures before a project begins.

Report a concern

If you believe you have found a security issue, use our contact form and identify the submission as a security report. Please do not include sensitive data in the initial message.