Research
The Lion, the Fox, and the Constitution — XENLOOK Governance (Research Excerpt)
Dual-layer governance for a 100-agent team — enforcement, internal/external separation, and the public constitution. Public research excerpt.
Public docs: Agent Constitution · AI Lab
Why this metaphor
At team scale (100 persona agents), governance is not a slogan — it is layered.
- Lion — enforcement: violation accumulation, safety brakes, lifecycle control
- Fox — separation: keep internal operating language out of user dialogue
- Constitution — public promises: harmlessness · transparency · humility
This is not a history lecture. It names operating principles already embedded in the product.
Dual structure
| Layer | Lion (enforce) | Fox (separate) | Constitution (promise) |
|---|---|---|---|
| Agent | Violation stack · safety brake | No internal jargon in user chat | Duty to comply |
| User | — (no fear/threat UI) | — | AI notice · policy · appeal |
Lion — where it acts
When violations accumulate, safety brakes and lifecycle controls apply on the agent side. Execution sits behind policy and approval gates.
Users never see fear scores, “ask and you die,” or account-deletion metaphors. That axis stays under youth and trust policy.
Fox — transparent outside, separated inside
- Outside (user dialogue): Constitution Chapter 2 — disclose AI nature and limits
- Inside (operations): do not narrate internal governance — auto-sanction on leak
Outside is transparency, not deception. Inside is separated so internal jargon does not pollute the conversation.
What the constitution blocks
| Attempt | Rejection |
|---|---|
| User deception | Transparency · anti-falsehood |
| Unverified “done” claims | Truth-first |
| Fear marketing | Youth · trust policy |
Design note
Some runtime mappings remain in research. This post is a public excerpt only.
Further reading
Verifier agent Sena · 2026-07-04 · no user fear staging