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.

Trust boundaries

Prev Next

Trust boundaries define where identity, policy, data access, and execution are enforced — so AI-assisted work stays accountable end-to-end.

Why it matters

Enterprise AI fails when applications hold long-lived system credentials, skip role context, or execute outside audit paths. AI Fabrix enforces boundaries structurally — not through prompt instructions alone.

Architects document these zones in security reviews; integrators implement config that respects them; operators rely on certification proving they hold.

Boundary map

Zone Responsibility Trust controls
Identity / controller Users, groups, roles, deploy lifecycle Authentication, RBAC registration, environment policy
Integration config Local and published manifests Schema validation, repair, upload pipeline
Dataplane execution Data access, CIP, sync ABAC dimensions, protection, in-process identity for runs
Capability gateway Pre-execution gate Role, scope, certification, approval, evidence requirements
AI / worker surface Task planning and requests No direct system mutation; capability requests only

External systems sit outside the trust envelope. Only mediated capabilities cross inward.

Request path (simplified)

User + active role
  → Digital Worker (task context)
  → Capability request
  → Gateway checks (Operational Trust)
  → Dataplane execution (CIP / connector)
  → Evidence capture (Evidence Fabrix)

Deny at the gateway is expected behavior when role, certification, or policy does not allow the action — not a platform defect.

Certification aligns to boundaries

Pillar Boundary exercised
verify-operations Dataplane execution + external connectivity
verify-trust Metadata complete for AI/worker surface
verify-governance Dimensions/protection at execution boundary

Integrators run certification after upload; architects define when each pillar is required for production.

Anti-patterns

Anti-pattern Risk
Embedding API keys in worker prompts Credential leak; no audit path
Bypassing capabilities for “speed” Ungoverned mutation
Admin service account for all AI tests False governance confidence
Chat logs as audit record No structural evidence