AI acts within limits it cannot exceed.

Chatur Systems separates interpretation from authority: models can understand intent, while platform rules decide what may actually happen.

Capability-scoped action

The AI can choose only among actions explicitly exposed to the interaction; it cannot expand its own access.

Verified important facts

Price, availability, payment state and access are read from and revalidated by deterministic services.

Idempotent execution

Retries are designed not to duplicate a booking, charge, message or other action.

Honest capability

An unconfigured provider or unavailable feature should be shown plainly, never presented as a success.

Policy input required

No invented human-escalation promise.

A universal “a person is always available” guarantee is not published because it is not verified across both products and every channel.

Principles need governance.

The legal framework still needs a named accountability owner and approved incident process.