Policy-by-proof is the enforcement model in which written policy defines the boundary of what an AI agent may do autonomously, and every action at that boundary must carry a proof of compliance with the policy that permitted it. Policy is not documentation of intent; it is the executable perimeter of autonomy.
In a policy-by-proof system, an operator can read — in plain language — exactly what the system is allowed to do without a human, change that boundary, and see the change take effect immediately. Autonomy is a setting owned by the accountable operator, not an emergent property of a model.
The proof half of the name is the enforcement mechanism: an action that cannot cite the policy clause permitting it does not execute. This converts policy compliance from an after-the-fact audit exercise into a precondition of runtime.
This vocabulary is implemented, not aspirational — it describes how the MAIA substrate actually runs. See the platform or read the research.