01
Capability ≠ authority
Being able to perform an action is not permission to perform it.
Governance
Control is a feature, not friction.
The most useful AI workforce is not the one with the fewest restrictions. It is the one that knows what it can do, what it should do, and when it must stop.
02
Validation ≠ permission
Confirming that an action is correct does not authorize executing it.
03
Intent ≠ authorization
A worker's plan is a proposal until a human with standing approves it.
04
Authority checks
Every consequential action is tested against the worker's declared authority before execution.
05
Evidence chain
Claimed completion is connected to obligations and independently established facts.
06
Exception handling
Missing, contradicted and stale results are surfaced rather than averaged away.
07
Approval paths
Consequential decisions route to a named human with the context required to decide.
08
Audit history
What was done, by which worker, under which authority, on which evidence.
09
Stop conditions
Workers are designed to halt and escalate when the situation exceeds their mandate.