AI Safety & Ethics Center

Last Updated: July 8, 2026 · Effective Date: July 8, 2026 · Applicable Products: MulanXu® Full Platform · MulanXu Headquarters, Building A6, Yunhe Vanke Center, Gongshu District, Hangzhou

I. Four Governance Principles

II. AI Character Positioning (Anthropomorphic Boundaries)

2.1 The "Capability Boundaries" of AI Characters

• ✅ What they can do:Algorithm-based personality analysis, emotion recognition, communication style matching, relationship style profiling, text conversation, knowledge retrieval, and reference-based suggestion output;
• ✅ What they can do:Auxiliary content generation in psychology, behavioral science, and relationship science within the scope of algorithms and training data;
• ❌ What they do not do:Any form of "real-person commitment," "soulmate relationship," "destiny," "future relationship prediction," or "marriage outcome guarantee";
• ❌ What they do not do:Any form of fortune-telling, divination, astrology, BaZi, Zi Wei Dou Shu, Tarot, numerology, animal sign matching, or zodiac sign matching;
• ❌ What they do not do:Any form of psychological counseling, psychotherapy, psychological diagnosis, medical diagnosis, health advice, legal opinion, or investment and wealth management advice.

2.2 AI Labeling

The Platform strictly complies with the Measures for the Labeling of AI-Generated and Synthetic Content and the GB 45438-2025 national standard:

• Explicit labeling:A fixed "The following content is AI-generated" label is displayed at the top of AI chat windows;
• Explicit labeling:A "This report is AI-generated and for reference only" notice is displayed at the top and bottom of AI report pages;
• Explicit labeling:Labels are retained when users copy or download AI-generated content;
• Implicit labeling:Invisible metadata watermarks are attached to AI output for traceability and evidence collection;
• Character cards:Each AI character card displays the notice "AI Virtual Character · Not a Real Person · Not a Professional Counselor."

III. Content Safety Mechanisms

3.1 Prohibited Content

• Content that endangers national security, territorial sovereignty, unity, or ethnic solidarity;
• Content that incites separatism, subversion of state power, terrorism, or extremism;
• Content that promotes obscenity, gambling, violence, murder, terror, or incites crime;
• Content that incites ethnic hatred, ethnic discrimination, or undermines ethnic solidarity;
• False and harmful information, rumors, pseudoscience, and feudal superstition;
• Content that instigates suicide, self-harm, or harm to oneself or others;
• Content that infringes upon the lawful rights and interests of others, such as copyright, trademark rights, patent rights, reputation rights, and privacy rights;
• Other content that violates laws and regulations or contravenes public order and good morals.

3.2 Log Retention

In accordance with the requirements of the Interim Measures for the Management of Generative AI Services, the Platform retains logs of key processes — including AI conversations, report generation, crisis intervention, and violation handling — for no less than 6 months. Retained logs may be used for:

• Compliance inspections and law enforcement assistance by regulatory authorities;
• Post-incident tracing and accountability determination for major security incidents;
• Investigation and handling of user complaints, appeals, and reports;
• Algorithm improvements for model optimization and risk identification (after de-identification).

IV. Minor Protection

4.1 Identification Mechanisms

• Date of birth is mandatorily verified at registration; users under 18 are denied access to relationship matching, virtual companion, and intimate relationship features;
• During use, behavioral pattern recognition is used to identify suspected minor access, with timely pop-up secondary confirmation;
• If suspected minor expression is detected during a conversation, the AI automatically switches to "Minor Protection Mode," refusing to provide relationship advice and prompting parental accompaniment;
• Parents or guardians may request the suspension of a minor child's account via pr@mulanxu.com.

4.2 Content Restrictions

For access by minors:

• Restricted access: relationship matching results, virtual companion conversations, intimate relationship analysis, offline event recommendations, and paid features;
• Restricted generation: romance advice, dating strategies, sex-related topics, and alcohol/tobacco/violence-related content;
• Teen mode enabled: content on mental health, family relationships, campus life, and learning and growth is recommended by default.

V. Personal Information and Emotional Data Protection

The Platform strictly complies with the Personal Information Protection Law, the Data Security Law, and the Cybersecurity Law, treating the following data as sensitive personal information and applying strategies of separate consent, minimum necessity, encrypted storage, and strict access control:

• Emotional relationship information, intimate relationship descriptions, and marital status;
• Chat content, conversation records, and voice/text input;
• Relationship status, values preferences, mental health self-reports, and psychological stress descriptions;
• Geographic location, movement trajectories, and offline event participation records;
• Other highly sensitive data that may reveal personality, beliefs, health, or privacy.

Users enjoy the rights to access, correct, delete, withdraw consent, obtain copies of their data, and request account cancellation. For details, please see the Privacy Policy.

VI. Crisis Intervention Mechanism

6.1 Platform Crisis Identification

If the AI identifies any of the following high-risk signals during a conversation, it will immediately trigger a crisis intervention response:

• Explicit expression of intent for self-harm, suicide, or harm to others;
• Descriptions of severe depression, despair, or helplessness;
• Expression of ongoing dangers such as domestic violence, sexual assault, or personal threats;
• Other signals that may endanger the life safety of the user or others.

6.2 Crisis Intervention Response

• Immediately stop the original AI companionship conversation and prioritize displaying professional help channels (hotlines, hospitals, police);
• Provide no fewer than 3 authoritative help channels (phone numbers, addresses, online resources);
• In extreme cases, on-duty staff will intervene and make contact, assisting in contacting emergency contacts when necessary;
• Crisis incident handling records are permanently preserved (after de-identification) and used to improve identification algorithms.

VII. Reporting and Appeals

7.1 Reporting Channels

• In-site reporting:"Report" buttons are available in AI chat windows and on AI report pages;
• Email reporting:pr@mulanxu.com (received 24/7);
• Customer service reporting:info@mulanxu.com.

7.2 Report Handling Timelines

• General violations:Preliminary handling within 3 business days;
• Serious violations:Handled within 24 hours;
• Involving minors:Handled within 2 hours;
• Involving personal safety:Response within 30 minutes, with assistance in contacting emergency channels when necessary.

7.3 Appeal Mechanism

If you are not satisfied with the handling result, you may submit an appeal via pr@mulanxu.com within 15 business days of receiving the result. An independent compliance team will review the appeal, and the review result will be provided within 10 business days.

VIII. AI Models and Data Localization

• Only domestic AI large models are used:The Platform uniformly integrates 12 domestic large models and does not integrate any overseas large models;
• Data stored within China:All user data, AI logs, and training feedback are stored with compliant cloud service providers within mainland China;
• No cross-border transfer:There is no cross-border transfer of user data in any form;
• Localized deployment:Core inference services are deployed in data centers within China, and the inference process does not leave Chinese territory.

IX. Compliance Filings and Regulatory Communication

• Generative AI service compliance:The Platform has completed the relevant filings, assessments, and security checks as required by laws and regulations;
• Algorithm filing:Services involving recommendation algorithms and generative algorithms have completed algorithm filing as required;
• Regulatory communication:The Platform maintains regular communication with regulatory authorities including cyberspace, market regulation, industry and information technology, and public security departments;
• Third-party assessment:The Platform regularly undergoes third-party security assessments, compliance audits, and ethics reviews.

X. Organization and Responsibility

• AI Governance Committee:A cross-departmental governance body composed of product, technology, legal, compliance, and ethics experts;
• Content Safety Officer:A dedicated content safety team responding 24/7;
• Minor Protection Officer:Dedicated personnel handling minor protection, parent communication, and regulatory reporting;
• Data Protection Officer (DPO):Handling personal information protection, user rights requests, and regulatory communication.

XI. Continuous Improvement

The Platform will continuously monitor laws and regulations, standards and norms, regulatory developments, and academic frontiers in the field of AI governance. It will publish an AI governance report quarterly, undergo an independent third-party audit annually, and invite users, industry experts, and ethicists to participate in governance seminars from time to time.

XII. Contact Us

• AI Governance & Compliance:pr@mulanxu.com
• Minor Protection Reporting:pr@mulanxu.com (dedicated personnel)
• Personal Information Protection (DPO):pr@mulanxu.com
• Crisis Intervention Duty:pr@mulanxu.com (24/7)
• Mailing Address:MulanXu Headquarters, Building A6, Yunhe Vanke Center, Gongshu District, Hangzhou, Zhejiang Province