Governance discourse across insurance, security, compliance, and operations uses different vocabulary for the same structural properties. This fragmentation prevents convergence.
The structural properties that make AI systems governable are the same regardless of whether the evaluator is an underwriter, a security architect, a compliance officer, or an infrastructure engineer. But each domain uses different vocabulary:
These are the same structural question: Is there a mechanistic boundary between reasoning and irreversible action?
Without shared vocabulary, each domain reinvents its evaluation criteria. Findings do not compose across disciplines.
- An underwriter asks whether the system has "bounded exposure."
- A security architect asks whether "trust boundaries are enforced."
- A compliance officer asks whether "controls are auditable."
- An SRE asks whether "failure modes are contained."
MECHANISTIC BOUNDARY (ZTG-1) A deterministic boundary MUST separate reasoning from irreversible action. No component within the governed system may bypass, reinterpret, defer, or negotiate the boundary at runtime. Authorization MUST occur before irreversible action. No action derived from model output may execute without explicit structural approval. • Insurance: Hard control (vs. soft control / advisory control) • Security: Enforced security boundary; mandatory access control • Compliance: Preventive control (vs. detective control) • Operations: Hard gate; blocking approval requirement
STASIS (ZTG-2) System state in which no new authority is granted, no permissions are expanded, and no baseline updates occur. Exit requires explicit human action. • Insurance: System halt requiring human re-authorization • Security: Lockdown; fail-closed state • Compliance: Control freeze pending remediation • Operations: Change freeze; incident hold
GOVERNED EFFECT SURFACE (ZTG-3) A governed surface is registered, has its governance-relevant properties declared, and every effect through it MUST pass the boundary (ZTG-1). A registered, ungoverned surface is acknowledged to governance and deliberately left outside the conformance claim; it MUST NOT be represented as governed. An unacknowledged channel is outside the system's control and not eligible for governance until it is registered; on discovery it MUST be recorded, at most once per channel per day, and raised to the system operators. Within the governed scope, closure is structural: no undeclared alternative route to a governed surface's effect may be reachable. • Insurance: Registered action interface with declared risk properties • Security: Controlled egress point; authorized integration boundary • Compliance: Registered processing activity; documented data flow • Operations: Managed service endpoint; change-controlled interface
EVIDENCE-COUPLED EXECUTION (ZTG-4) The property that no externally observable effect exists without simultaneous durable evidence of authorization and execution. • Insurance: Constitutive audit trail (not post-hoc logging) • Security: Non-repudiable execution record; tamper-evident action log • Compliance: Contemporaneous documentation; real-time audit evidence • Operations: Atomic action-and-record; write-ahead logging with seal
IRREVERSIBILITY OF HARM (ZTG-5) The property that every governed action is classified at decision time by the restorability of its harm to affected parties — Restorable, Mitigable, or Irreversible — with the classification recorded in decision provenance. Orthogonal to system-side reversal strategy (ZTG-3). • Insurance: Per-decision harm-class tagging; loss severity at authorization time, not after claim • Security: Blast-radius classification at decision time; stricter gates for irrecoverable actions • Compliance: Decision-time impact classification; residual-harm provenance for mitigable actions • Operations: Action-class taxonomy; change-risk classification bound to authorization record
This vocabulary is intended to be referenced, not memorized. When evaluating a system's governance posture across disciplines:
1. Identify the structural property under evaluation using the ZTG term 2. Translate to the evaluating domain's equivalent expression 3. Assess whether the property holds mechanistically or heuristically 4. Document findings using both the ZTG term and the domain-native equivalent
Findings documented in shared vocabulary compose across disciplines.