AI → Signals → Governance → Authorization Decision → Execution Runtime → Attestation
What humans control
Humans define authority through:
signed policies
authorization rules
execution constraints
These policies determine what is allowed or denied.
What AI does NOT control
AI systems do NOT:
approve execution
define policies
override governance
execute actions directly
AI is limited to generating signals.
Governance enforces human authority
Governance evaluates:
verified signals
signed human policies
It produces:
Authorization Decision
This decision reflects human-defined authority.
Execution follows authority, not intelligence
Execution is only allowed after authorization:
Authorization Decision → Execution Runtime → Execution
This ensures AI cannot bypass human control.
Why separation matters
If AI had authority:
decisions would be non-deterministic
systems would be non-auditable
execution could not be verified
Separating intelligence from authority ensures system safety.
Key invariant
Authority is defined once by humans
Authority is enforced deterministically by governance
Authority is never inferred by AI
Summary
Parmana ensures:
AI generates signals (intelligence layer)
Humans define authority (policy layer)
Governance enforces authority (decision layer)
Execution Runtime enforces decisions (execution layer)
Attestation proves correctness (verification layer)
