Not all agents are agents.
The word “agent” is overloaded. A summariser that drafts an email and a payment system that moves your money are both sold as “AI agents” — yet one carries no consequence and the other is irreversible. The KYE Consequential Actor Framework™ grades an actor across five levels of consequence, so you govern by how much finality an action requires — not by the marketing label. KYE Protocol™ is the trust layer for consequential action.
Why a consequence axis
You cannot govern a thousand “agents” the same way and stay both safe and shippable. A CISO who applies banking-grade controls to a chat assistant burns the budget that a claims agent actually needs. The KYE Consequential Actor Framework™ separates the question “what is it called?” from the question that matters: “how consequential is the action, and how final must it be?”
Consequence sits across two governance axes KYE™ already defines, and crosswalks onto both:
- Regulatory risk — the EU AI Act™ floor (minimal, limited, high). The framework names the typical risk class each actor level falls into.
- Enforcement depth — the KYE Conformance Maturity Ladder™ (L1–L5), from schema-valid to reality-coupled. Each actor level maps to the conformance level it requires.
- Actor consequence — the new axis here: how final an action must be, from none to mandatory-with-contestability.
The five consequential actor levels
Each level says what the actor is, gives concrete examples, states the governance it requires, and maps to the conformance level on the KYE Conformance Maturity Ladder™ from constitution.9. Read down the table to see consequence — and therefore governance — rise from transparency to mandatory contestability.
| Level | What it is | Examples | Governance it requires | Conformance |
|---|---|---|---|---|
| L0 Informational | No finality — produces information a human reads; never completes a consequential act. | Content generators, chat assistants, summarisers. | Transparency-grade: identify the actor, log the interaction. Risk class minimal. | L1 — Schema-valid. |
| L1 Advisory | Low finality — recommends; a human still decides and acts. | RAG assistants, recommendation engines, copilots. | Transparency plus delegation capture — EU AI Act™ Article 50 disclosure; the basis of each recommendation recorded. Risk class limited. | L2 — Entity + delegation captured. |
| L2 Operational | Scoped finality — acts within bounded, reversible workflows under delegated authority. | Workflow automations, multi-step orchestrators. | Action Admissibility™ enforced at runtime before each act, fail-closed — not reconstructed after. Risk class limited. | L3 — Admissibility enforced. |
| L3 Consequential | High finality — moves money, decides claims, places trades, affects health or legal outcomes. | Claims, payment, medical-triage, and trading agents. | Authority Finality™ — the decision is HSM-signed and replay-derivable from public keys alone, with human review for irreversibles. Risk class high. | L4 — Finality replayable. |
| L4 Systemic | Mandatory finality with contestability — consequence crosses organisational and jurisdictional boundaries. | Cross-organisational settlement actors, high-risk autonomous systems. | Contestability and jurisdiction rails — the revocation cascade, contestability, and reality-coupling drift wired into runtime effect. Risk class high. | L5 — Reality-coupling + contestability + revocation. |
The crosswalk is canonical: every actor level resolves to one rung of the KYE Conformance Maturity Ladder™ and one tier of the EU AI Act™ risk floor. You classify an actor once; the governance follows deterministically.
Finality is the dividing line
The jump that matters for you is L2 to L3. Below it, an action is reversible or merely advisory; above it, an action is consequential and hard to undo — so it demands Authority Finality™. At L3 the word “agent” stops being a label and becomes a regulated actor whose decision must be sealed, signed, and provable to a third party. The proof is a signed Evidence Pack™ that replays from public keys alone — no trust in KYE Protocol™ required.
- You hold the classification. Grade each actor once against the five levels; the required conformance level and risk class follow from the crosswalk.
- Governance is proportionate. A summariser is not policed like a payments agent — budget and controls land where consequence actually is.
- Proof, not assertion. From L3, every decision is replay-derivable from public keys, so your auditors and regulators verify it without trusting the vendor.
What KYE™ does — and does not — do
KYE Protocol™ governs actors as first-class principals: it resolves their identity, binds their authority, meters their use, and seals their evidence. It does not run an agent framework. If your team builds on LangChain, CrewAI, the OpenAI Agents SDK, or the Claude Agent SDK, KYE™ governs the consequential actions those agents take — it never replaces the framework that runs them.