Security and Trust
Thyris treats security as part of product and integration design. Controls are applied at data, identity, tool, workflow, and operational boundaries rather than relying on model behavior alone.
Core principles
Minimize data exposure
Only the fields required for a defined task should cross an integration boundary. Sensitive content can be detected, redacted, transformed, or blocked before it reaches an AI provider or downstream tool.
Keep secrets out of content
Credentials belong in an approved secret-management mechanism, not prompts, source files, examples, logs, or client applications. Integrations should use separate credentials and access scopes for each environment.
Separate customer boundaries
Stores, realms, users, sessions, and environments must be resolved from trusted identity context. A caller-provided identifier is never sufficient evidence that the caller may access the corresponding resource.
Control write operations
Operations that change state should validate authorization and business rules at execution time. High-impact actions should use explicit confirmation, idempotency protection, and a clear record of the requested operation.
Preserve evidence
Operational logs and audit events should make it possible to reconstruct important requests, policy decisions, tool calls, and failures without unnecessarily recording sensitive payloads.
Customer responsibilities
Customers define which data is sensitive, who may access each capability, which actions require approval, how long operational records are retained, and how incidents are handled within their environment. The product-specific security guides describe the controls available for implementing those decisions.