Application security
- Authorization enforced on the server for every resource and action; hidden UI controls are never relied on.
- Input validation at trust boundaries and parameterized queries.
- Rate limits, abuse detection and provider quotas.
- CSRF protection, secure cookies, strict security headers and a restrictive content security policy.
- Webhook signatures, timestamp validation and replay protection.
- Outbound network destinations restricted to an approved allowlist.
AI-specific safeguards
- Tool allowlists with schema validation and server-side authorization.
- Retrieved and user content treated as untrusted data; prompt-injection tests in evaluation.
- Minimization of sensitive data sent to model providers.
- No training on customer data without explicit, separate agreement.
- Human review for low-confidence or high-impact outputs.
Secure delivery
- Code review, automated tests and CI gates on every change.
- Dependency, container and secret scanning; software bill of materials for releases.
- Separate environments and credentials for development, staging and production.
- Threat modeling and security review for high-risk features.
- Release gates that block promotion when isolation or authorization tests fail.
Resilience and incident response
Backups are verified by restore tests, not only by job success. Recovery objectives are defined per tier. Incident procedures cover suspected cross-tenant access, secret exposure, data loss and harmful automation, with customer notification according to your agreement.
What we do not claim
This page describes controls and design intent. It is not a certification. Where we hold or pursue independent attestations such as SOC 2 or ISO 27001, we will list them here with scope and dates. Security needs also depend on how you configure and use the platform.