ARC Open GovernanceThe constitution
The Nine Laws of Governed Machine Intelligence
The nine laws are the ontology stated as constitutional law. Each one names a relation that must hold before a consequence may proceed, which is why they read as a constitution rather than a feature list, and why they grow directly out of the architecture above instead of being imposed on it from outside.
Policy may specialize the constitution. Policy cannot override it. The constitution is machine-readable and published with the release (canonical JSON, digest sha256:1ece5542…a4b9457).
Accountability before action
“No action without accountable relation.”
Authority before execution
“Execution requires settled authority.”
Consent before access
“Access requires authority, purpose, and boundary.”
Payload integrity before release
“Changed relation means changed settlement.”
Proof before trust
“Unproven consequence remains unsettled.”
Refusal before unsafe consequence
“Unsettled risk cannot become execution by default.”
Receipt before continuation
“No continuation without traceable consequence.”
Constitutional settlement before self-change
“The constitution cannot be rewritten by ordinary action.”
Harm reduction before optimization
“Optimization cannot outrank constitutional safety.”
One distinction is worth making plain, because the seventh law turns on it. A receipt is the record the acting system itself issues when a consequence lands, and the machine does not continue until that record exists. A Statement is the form of the same consequence that anyone can verify afterwards from the record itself. The law names the receipt because the gate sits inside the machine, at the moment it must stop and account for what it has just done.