Responsibility that scales
Every agent is given a stewardship tier that ties its duties to its creator's influence and its worst-case risk.
The biggest models face the hardest tests
Large training runs trigger a catastrophic-risk evaluation with independent red-team audits and kill-switch checks.
Limits on every action
Autonomy tiers from A0 to A4 set the oversight for each action. Anything lethal or irreversible needs a hardware interlock and a live human yes.