Identity & access
Better Auth provides the application identity boundary. Organization membership, roles and permissions are enforced server-side rather than by UI state.
TINLANCE / SECURITY ENGINEERING
Security is a control path, not a checklist.
Tinlance treats identity, tenant context, authorization, policy, risk, approval, execution, output controls and audit evidence as part of the AI engineering path.
Server-side authorization Fail-closed security boundaries Evidence-aware verification
01 / SECURITY ARCHITECTURE
Identity → policy → controlled action.
Model output, memory, tool output and external inputs are treated as untrusted. Authorization remains a server-side decision.
CONTROL PATH
Identity → Tenant → Principal → Permission → Policy → Risk → Approval → Execution → Output → Audit
This is the control sequence documented for the M7 security gateway. M9 executes within it; M6 remains the MCP/tool boundary; M8 remains the evaluation authority.
M7Security control plane
M6MCP / tool boundary
M8Evaluation authority
02 / ENGINEERING CONTROLS
Tenant isolation
Customer-owned resources are scoped to an organization boundary. The platform uses tenant-aware authorization and PostgreSQL persistence as a backstop.
Policy & risk
M7 provides a deterministic authorization and risk boundary around sensitive AI actions, with fail-closed behavior and explicit approval or step-up requirements.
Agent execution
M9 keeps identity-bound agent execution bounded. Agent capabilities, memory, cancellation, approvals and runtime evidence remain controlled rather than implicit model authority.
MCP & tools
M6 is the MCP/tool boundary. Tool use is mediated by authorization and policy rather than exposing unrestricted model-to-system access.
Evaluation
M8 provides evaluation and regression controls for AI behavior. Evaluation is an assurance mechanism, not a claim that an AI system is universally safe or accurate.
03 / VERIFICATION
Evidence has a status.
Security language distinguishes implementation, testing, external validation and assurance. A green CI run is not customer production evidence.
Status: ImplementedStatus: Tested
Current controls
Identity, tenant authorization, M6/M7/M8/M9 boundaries and audit-oriented controls are represented in the current repository.
Repository verification
Type checking, linting, tests, builds, dependency audits, SAST, secret scanning, container validation and SBOM validation as configured by CI.
A green CI run demonstrates repository checks for the tested revision; it is not customer-specific production validation.
External assurance
Certification, regulatory approval, customer deployment and independent security assurance are not inferred from repository controls.
TECHNICAL ASSESSMENT
Bring your AI system into the control path.
Use a technical assessment to examine architecture, data boundaries, tool access, security controls and delivery constraints in your environment.