Documentation Index

Fetch the complete documentation index at: https://docs.aifabrix.ai/llms.txt

Use this file to discover all available pages before exploring further.

Authentication for trust

Prev Next

Operational Trust depends on knowing who is acting and how credentials are managed — without exposing secrets to AI clients or duplicating identity in every integration.

AI Fabrix works with the organisation’s existing Azure controls, including Entra ID for identity and Key Vault for secrets. It does not replace the identity provider or create a separate user identity store.

Why it matters

Enterprise systems use many authentication models: API keys, OAuth, certificates, and platform-managed credentials. Architects need a trust lens — which method fits which system class, where secrets live, and what AI never sees — separate from integrator JSON editing.

If authentication is treated as “just connect the API,” teams store keys in prompts, share tokens in chat, or grant AI broad service accounts. Operational Trust blocks that pattern: humans and platform services authenticate; Role Assistants request capabilities, not vendor credentials. AI has no authority of its own — every action runs for an identified person.

How it works

Enterprise identity (users, roles)
  ↓
Platform credential store (encrypted, audited)
  ↓
Integration authentication (method + secret reference)
  ↓
Governed capability execution (no raw secrets to AI)
Concern Trust boundary
Human login Corporate identity — who may administer and who may use Role Assistants
Caller authority for AI Fabrix APIs Real-user authority — not an autonomous system actor
Vendor / Connected System credentials Encrypted store; referenced by integrations, never pasted in prompts
Platform-managed downstream Platform acquires and injects downstream tokens where configured — still not a substitute for real-user authority on business work
AI access Capability keys and metadata — never vendor API keys

Declare method on the Connected System, bind secrets through the platform vault, validate before upload, and re-certify when credentials rotate. Integrators confirm current methods on the authentication methods catalog.

When identity cannot be verified, Operational Trust stops safely — cannot verify means do not execute.

When architects choose what

System class Typical pattern Trust note
SaaS with private app / API key API key in vault Rotate on offboarding; scope to sandbox for pilots
Enterprise OAuth OAuth client + secret in vault Prefer dedicated integration app registration
Platform-owned shells / facades Platform-managed downstream auth No vendor secrets on the slot; business calls still require real-user authority
Microsoft / Entra workloads Certificate or Entra Align with corporate identity governance
Public read-only APIs No vendor credential True anonymous downstream only — not “forward the user token”

Architects document who may create credentials, rotation cadence, and which environments share secrets (dev vs prod never share keys).

Example

A CRM integration uses an API key stored in the platform vault. Sales Managers authenticate to AI Fabrix with corporate identity; the Sales Assistant requests customer.list — the platform executes with the integration credential. The assistant never receives the CRM key; audit logs show user, role, capability, and outcome.

Common mistakes

Mistake Risk Mitigation
Shared service account for all AI use No user-level accountability Role-scoped subjects in governance certification
Keys in environment files on laptops Leakage, drift Vault references in manifests only
Skipping re-validation after rotation Silent production failures Include auth changes in release checklist

Business value

Clear authentication boundaries make AI programs auditable: identity proves who, credentials prove system access, and capabilities prove what AI may request.

Limits

Supported authentication methods and vault key patterns evolve with platform releases — integrators confirm current methods on the authentication methods catalog before production rollout.

Corporate policies may forbid specific methods (for example long-lived API keys) even when vendors allow them — this page does not replace enterprise security standards.