AI Use Policy

Effective date: October 12, 2026

KOncierge Corp. (주식회사 콩시어지, the "Company", "we", "us") publishes this AI Use Policy (this "Policy") to explain how we use AI in the KOncierge Concierge Agent, what information reaches the AI, and how people stay involved. Matters of personal data processing not covered here are governed by our Privacy Policy.

1. Overview at a Glance

This table is a summary only; the full sections below are definitive.

2. Services That Use AI

2.1. We use AI in the KOncierge Concierge Agent (the "Agent"), which we provide in LINE, WhatsApp, and WeChat conversations connected through Channel Talk.

2.2. In the Agent, AI is used to:

• generate replies to your messages, such as Korea travel information and place recommendations;
• prepare restaurant booking requests you ask for, so they can be passed to our staff;
• summarize earlier conversation so later replies keep the context;
• translate your messages related to a restaurant booking request into Korean so our staff can read them; and
• draft the progress notices our staff send you about a booking request.

3. AI Model Provider

3.1. The Agent uses language models from OpenAI through its API.

3.2. When you ask for place recommendations, search terms written by the AI are sent to Google Maps place search (Google Places) to verify places.

3.3. How the AI model provider processes conversation content, and the restriction on using it to train models, are governed by Sections 5 and 7.2 of the Privacy Policy.

4. Information Sent to the AI

4.1 Information sent

• The message you have just sent
• Some recent messages from the same conversation (up to 20, including messages from our staff) and the summary of earlier conversation that we store
• Your open restaurant booking requests (restaurant, date, time, party size, status) and our staff's questions and notes about them
• The conversation language

4.2 Information masked before sending

Messages used to generate replies and conversation summaries are sent to the AI after email addresses, phone numbers, card numbers, passport-number-like strings, numbers of 8 or more digits, and member-link codes are masked. Masking is automatic and pattern-based, so we cannot guarantee that every identifier is masked.

When we translate messages related to a restaurant booking request into Korean for our staff, however, the original message text is sent unmasked so that the translation is accurate. Please avoid sharing more personal detail than your request requires; in particular, do not send government ID numbers, full passport or card numbers, or passwords in chat.

4.3 Information not sent

• Account and profile data we hold (name, email address, phone number, and the like) and messaging profile data. Anything you write in the conversation yourself is handled under 4.1 and 4.2.
• Payment and transaction data, and whether you hold an access pass
• Photos, files, and other attachments. Only the fact that an attachment exists and its file type are sent — never the file itself or its address.

5. What the AI Does Not Do on Its Own

5.1. The functions available to the AI are limited to place search, replying to you, requesting a handoff to staff, and passing restaurant booking requests, changes, and cancellation requests to staff.

5.2. The AI does not, on its own:

• Confirm bookings. The AI only passes restaurant booking requests to our staff, and says a booking is confirmed or cancelled only when the status set by our staff shows it. Other bookings, and creating, changing, cancelling, or confirming arrangements our staff manage, are handed off to staff.
• Process payments or refunds. Payment and refund requests are handed off to staff.
• Set prices. The AI has no function to set or change prices.

5.3. Every AI reply is checked automatically before it is sent to you. A reply that fails the check — for example, one that names a place not confirmed by place search, our reference materials, or the conversation, or says a booking is confirmed when it is not — is not sent, and the conversation is handed off to staff instead.

6. Handoff to a Person

6.1. A conversation is handed off to our staff when:

• you ask for a person or raise a complaint;
• your request is one our staff must handle under 5.2;
• the situation appears to be an emergency, in which case the Agent first gives Korea's emergency numbers (119, 112, 1330); or
• an AI reply fails the check in 5.3 or cannot be generated normally.

6.2. Once a conversation is handed off, the AI does not reply in it. We alert our staff through Channel Talk and our internal work messenger, and they reply after reviewing the conversation history and summary.

6.3. When our staff finish, the conversation may return to AI replies.

7. Retention

7.1. Agent conversation records (messages, records of how AI replies were generated, conversation summaries, and restaurant booking requests) are stored in our database and deleted automatically 180 days after the conversation's last message.

7.2. If you delete your account, Section 9.5 of the Privacy Policy applies.

7.3. Copies of conversations held by messaging platforms such as LINE, WhatsApp, and WeChat are governed by each platform's own privacy policy and are outside the scope of our deletion.

8. Limitations of AI

AI replies are generated automatically and may contain inaccurate or outdated information or misunderstand what you mean. Please confirm important details — such as opening hours, prices, and availability — with our staff or the business concerned.

9. Your Rights and Contact

9.1. You may at any time ask to talk to a person instead of the AI, and object to an AI reply.

9.2. Your rights over personal data, and how to exercise them, are set out in Section 12 of the Privacy Policy.

9.3. Please send questions about this Policy to our Privacy Officer (partnership@koncierge.co.kr).

10. Changes to This Policy

We may update this Policy to reflect changes in how we use AI, our model providers, or applicable law, and will announce updates through the Services at least 7 days before they take effect. The effective date at the top shows the current version.

11. Language

This Policy may be published in multiple languages for convenience. In case of inconsistency, the Korean version prevails.