SOC 2 Type 2 certified, HIPAA and HITRUST compliant, and single-tenant for every customer, with data protected in transit, at rest, and in use.
SOC 2 report, security policies, and subprocessors.
SOC 2 Type 2 certified, with three consecutive audits completed. HIPAA, HITRUST, FERPA, COPPA, and CIPA compliant, with business associate agreements available for healthcare customers. Built to support GDPR and CCPA. ISO 27001 certification in progress.
Every customer runs on its own single-tenant service in a dedicated AWS environment. Services, data, search indexes, logs, and encryption keys are isolated to that environment and never shared with another customer. Liminal personnel can access a customer environment only to troubleshoot, through MFA-protected accounts with break-glass approval.
US, UK, and EU data residency available, with US hosting by default.
Liminal detects PHI, PII, PCI, and intellectual property in prompts and files, and protects it according to your policies before any prompt reaches a model. Detection models are trained only on synthetic data.
Customer data is encrypted in transit and at rest. Prompt logs are protected with AES-256 encryption, credentials for connected services are encrypted with AWS Key Management Service, and encryption keys are isolated to each customer's environment.
Liminal never trains on customer data. Requests to model providers are sent only after sensitive data is handled according to your policies. Liminal holds business associate agreements with Anthropic, OpenAI, and Google for the models it provides; Perplexity and xAI models aren't covered, and administrators can choose not to enable them.
Single sign-on through SAML 2.0 with any compliant identity provider, including Microsoft Entra ID, Okta, and Ping Identity. Role-based access control with custom roles.
Every interaction is logged with the original and protected prompt, user, model, and policies applied. Logs can be downloaded as CSV, JSON, or XML, or exported on a schedule to AWS, Azure, or Google Cloud storage for your SIEM. Email alerts flag sensitive data as it's detected.
An independent third party penetration tests the Liminal platform at least once a year. Results are available in the Trust Center.
Liminal maintains a formal incident response plan, reviewed at least annually. If an incident involves unauthorized access to customer data, affected customers are notified within 72 hours of confirmation, as set out in the Data Processing Addendum.
Prompt and activity logs are retained for the life of your contract, and your administrators can delete them on any schedule. Sensitive values are held only long enough to restore them in the response. When a customer leaves, Liminal deletes all customer data and decommissions the environment within 30 to 90 days.
All Liminal subprocessors are located in the United States. The current list is available on the Subprocessors page.
Tell us about your requirements and we'll walk your team through Liminal's security and compliance.
