AI Notice
AI Usage Notice
AI Framework Act §20 · advance notice to users
The short version
dadokim grounding cards use generative AI for card generation and safety classification.
Where is AI used?
- Safety classification (Safety Router) — automatically detects crisis signals in what you type (suicide · self-harm · immediate danger). An LLM runs a second-pass classification that complements 100 hardcoded regex patterns.
- Card generation — when none of the 55 beta cards fits, one new card is generated. Every generated card passes through a wording filter (Medical Service Act · Pharmaceutical Affairs Act) before it’s shown.
Is this “high-impact AI”?
No. This service is not used for decisions in healthcare, finance, credit, hiring, education, or the public sector. It’s a small ritual for your mind — comfort content. It doesn’t diagnose, treat, or prescribe anything.
That said, because emotional well-being sits close to healthcare, we refresh a self-assessment memo every quarter. The current memo is kept internally, and a public summary is available on request.
Is my information sent to the AI?
- The one line you type (max 120 characters) is sent to an LLM API provider (US).
- It’s discarded from memory as soon as your card comes back — not stored (no raw text, no hash).
- Your name, email, phone number, and other identifying details are never sent to the AI.
- The provider does not use your input as training data (LLM API policy).
International transfer · subprocessing of personal data (PIPA §28 · §17)
To use your one free-text line for card generation, the following subprocessing and international transfer take place.
| Item | Details |
|---|---|
| Subprocessor | An LLM API provider (US) |
| Destination country | United States (US) |
| Transfer method | HTTPS · TLS 1.3 encrypted transport · LLM API |
| Data transferred | Your one free-text line (max 120 characters · no identifying info) · category · safety level |
| Purpose | Generating one grounding card (LLM inference) |
| Retention | Provider side: auto-deleted after 30 days (LLM API policy) · dadokim side: discarded instantly — not stored |
| Used for training | No (the provider’s commercial terms — no training on customer data) |
| Right to refuse | If you decline the international transfer, you can’t use the service (the LLM is core to how it works) |
※ The LLM API provider’s data-processing terms commit to no training on customer data and automatic deletion after 30 days.
Disagree with something a card said?
- Use the “This card isn’t right” button on the card result screen to report it on the spot
- Email: privacy@dadokim.com
- Reported cases feed into a red-team analysis every 90 days
The safety net (4 layers)
- 100 hardcoded keywords (≤10 ms · instant)
- LLM classification (1–2 s · complements the hardcoded list)
- Pre-output wording filter (Medical Service Act · Pharmaceutical Affairs Act · one regeneration retry · safe-card fallback on failure)
- User reports (4th layer · fed back in)
When Level 2–3 signals are detected (a direct mention of suicide or self-harm · immediate danger), card generation itself is blocked and you’re routed toemergency guidance (in Korea: 109 · 119 · 1577-0199 · outside Korea: findahelpline.com).
※ The card generator is Korean-only for now, and automated crisis detection currently covers Korean text only — the English safety router isn’t live yet.
Related laws
- Framework Act on the Development of Artificial Intelligence and the Establishment of Trust (Korea’s “AI Framework Act” · in force since 2026-01-22)
- Personal Information Protection Act (PIPA)
- Medical Service Act · Pharmaceutical Affairs Act · Mental Health Welfare Act (wording filter)
- E-commerce Act (business information · payments · refunds)
Self-assessment memo (summary). Refreshed once a quarter (March · June · September · December).
- v1.0 (2026-05-27): initial draft · legal review pending
- v2.3 (2026-05-28): incorporated counsel notes v0.1 · expanded the international transfer and subprocessing details
- Next review: 2026-08-27