● Trust & security
We test your AI. We don't keep it.
A safety vendor that leaks is worse than no vendor. These are the commitments in every contract.
- Training
- Customer prompts, outputs, traces, datasets and findings are never used to train our models and never shared with other customers.
- Deployment
- Managed SaaS (US or EU), single-tenant, your VPC on AWS, Google Cloud or Azure, or air-gapped on-prem.
- Retention
- Configurable per workspace. Guard supports zero retention: decisions are logged, payloads are not.
- Encryption
- TLS 1.2+ in transit, AES-256 at rest, customer-managed keys on Enterprise.
- Access
- SSO and SCIM, role-based access, just-in-time production access for our staff with full audit trails.
- Red-team data
- Findings and exploit transcripts are stored in isolated, access-logged projects and deleted on request at engagement close.
- People
- Background checks and NDAs for all employees and specialists. Annual security and data-handling training.
- Compliance
- Our security controls and subprocessor list are available under NDA. Contact us for our security questionnaire responses and DPA.
- Disclosure
- Found a vulnerability in our products? Contact us through the security option on the contact form. We respond within 48 hours.
Next step
Find out what your model does before your users do.
A 45-minute briefing with our evaluation team. Bring a use case; leave with a threat model, a draft eval plan and a sense of where you stand.