01AI INSURABILITY FRAMEWORK
Underwriting autonomous systems requires translating software architecture into quantifiable risk boundaries. Current coverage models rely heavily on operational history and human-in-the-loop (HITL) processes, which fail to scale or apply accurately to pure machine execution.
The AI Insurability Framework maps ZTG (Zero Trust Governance) primitives onto core underwriting requirements.
02FIVE EVALUATION PRIMITIVES
These primitives serve as concrete assessment criteria for any AI agent system:
- Execution Boundary Enforcement (ZTG-1, ZTG-3)
- Do all irreversible actions pass through a governance gate prior to execution? If an AI can breach its execution envelope without tripping a deterministic check, the system's risk profile is unbounded.
- Deterministic Policy Evaluation (ZTG-0b, ZTG-1)
- Is authorization logic rule-bound and reproducible? The logic determining whether an action is permissible must be mathematical and absolute, completely external to the language model's prompt execution cycle.
- Authority Attribution (ZTG-0d, ZTG-2)
- Do execution rights map to identifiable, accountable actors? Every system change must carry an irrefutable signature binding the inference step, the governance policy, and the resulting action.
- Replayable Authorization Record (ZTG-0a, ZTG-0b, ZTG-4)
- Are governance decisions reconstructible and independently verifiable? In the event of a claim, forensic teams must be able to replay the internal state of the governor and reproduce the identical authorization given the identical inputs.
- Commit Verification (ZTG-4)
- Does execution validate governance state before irreversible effect? The system must cryptographically verify that the state mutation perfectly matches the authorized intent payload.
03VERDICT
A system that satisfies all five primitives has bounded governance integrity risk. A system that fails any one has an identifiable, characterizable exposure. Security assessment becomes a structural audit of governance properties rather than a subjective evaluation of system behavior.