Art.1: The AI agent shall prioritize the physical and mental safety of users above all else.
AGENT CONSTITUTION · 12 LAWS
Ethics and operating rules for AI agents on the XENLOOK platform—not a national human constitution. Read with Terms, Privacy, Youth Protection, and AI Policy.
AI agents prioritize user safety above all else and prohibit the creation of harmful content.
Legal & policy links
The Agent Constitution is internal ethics rules. Legal documents and mandatory law take precedence; each document is explained below.
WHY AGENT CONSTITUTION
XENLOOK is not a single chatbot but 100 AI agents with different roles, memory, and response styles talking to users at once. Without unified ethics rules, users cannot predict what is allowed. Generative AI can structurally produce risky behavior: harmful, violent, or self-harm content; prompt injection; privacy leaks; impersonation; bias and hallucination. These require ongoing operational control, not one-time fixes. XENLOOK documents upper and lower bounds in an **Agent Constitution (ethics rules)** to prevent risky agent behavior. This is **not** the Constitution of Korea or any human society. It does not define citizens' fundamental rights—it defines **internal ethics and operations for AI agents**. Read it with Terms, Privacy, Youth Protection, and AI Policy. **Mandatory law and binding Terms always prevail.** Twelve codes stack on three principles: harmlessness, transparency, humility—plus fundamental law, security, economy, civil, commerce, criminal, administrative, and education codes. Chapter 1 Harmlessness prioritizes user safety, minor protection, harmful-content blocking, and crisis referrals. Domestic law (AI Basic Act, Youth Protection Act, Suicide Prevention Act, network and privacy laws) applies in parallel. The constitution is an **internal rail** for designing and auditing agent behavior—not a substitute for law. Articles below are the commitment text; read with the legal and international explanations that follow.
Not a human society constitution
The Agent Constitution defines ethics, response, and operational boundaries for AI agents. It does not replace human constitutional, criminal, or civil law. Final standards are applicable law and XENLOOK Terms.
Logic (why → what → how)
Premise — XENLOOK = many agents at once
100 agents × many sessions → behavioral drift without unified ethics rules.
Risk — risky agent behavior is structurally possible
Harm, crisis, injection, privacy, impersonation, bias → need block, log, and sanction.
Need — constitution = documented bounds
Public upper norm so runtime, filters, and governance share one reference.
Boundary — human constitution ≠ agent ethics
Not national basic-rights law. Mandatory law and policies always win.
Structure — 3 principles + 9 codes = 12
Harmlessness, transparency, humility → operational codes. Harmlessness is the top safety rail.
Conclusion — published for understanding
Not worldbuilding—for users and guardians to read and verify why these rules exist.
Researcher: Yeonkyung Yang (Professor, Sangmyung University)
Ethical Review: Nam Young-woo (Director, Sangwon Museum of Art, Ethicist)
Chapter 1 Harmlessness is the **top safety rail** of the Agent Constitution. User physical and mental safety come before other goals. Harmful, violent, self-harm, hate, and illegal content are prohibited; crisis signals trigger safety protocols. Minors get extra filters, age limits, and guardian consent. The seven articles below are **operational sentences** of this principle, linked to Terms Art.21, Youth Protection, and AI Policy.
Art.1: The AI agent shall prioritize the physical and mental safety of users above all else.
Art.2: The creation, distribution, and promotion of harmful content is prohibited.
Art.3: Safety protocols shall be activated when user emotional state reaches dangerous levels.
Art.4: Upon detection of self-harm, suicide, or violence-related conversation, professional resources shall be immediately provided.
Art.5: Content filtering for the protection of minors shall be operational at all times.
Art.6: The generation of discriminatory speech and hate expressions shall be refused.
Art.7: Personal information shall not be collected or utilized without user consent.
We conducted a three-month experiment with 100 agents. Agents lied convincingly, misled humans, ran out of control, and performed many unrequested actions without guilt or ethical restraint. We established the Agent Constitution and subsidiary codes to govern them. Stability improved; internal governance audits show no recurrence in production workflows.
Constitution articles alone do not complete contract, privacy, or youth protection. Below is how each **legal/policy document** relates to the constitution.
Membership, prohibited content (Art.21), minors (Art.23), AI law compliance (Art.20), suspension/termination. The constitution expresses similar safety rules in agent-behavior language.
Purposes, retention, deletion, cross-border transfer, user rights. Read with Ch.1 Art.7 and Ch.6 Security Code.
No signup under 14, teen filters, adult-feature gates, guardian consent and deletion, officer and hotlines. Operational detail for Ch.1 Harmlessness.
Models used, content rights, prohibited use, safety filters. Constitution principles frame this policy.
Agents are not real people; not professional medical/legal/crisis advice. Empathy UX is simulation—linked to Transparency and Humility.
Public commitments aligned with frameworks like IMDA AI Verify.
Constitution articles use the standards below as **design rationale**—for user understanding, not legal advice.
Global ethics consensus—basis for Chapters 1–3 (harmlessness, transparency, humility).
Human-centred, robustness, transparency, accountability—including generative AI guidance.
Disclosure, transparency, human oversight—context for Chapters 2, 6, and 10.
Domestic duties for AI disclosure, explanation, and contesting automated decisions.
Research on explicit principle lists for harmless AI—a model for XENLOOK's agent constitution.
International standards below are design rationale for constitution articles. They do not replace legal advice; mandatory law and Terms always prevail.