How Mitra Protects Your Child

Our commitment to safety, transparency, and your family's rights under Korean and international law.

1. We know the law, and we welcome it

South Korea's AI Basic Act came into effect on January 22, 2026, making Korea the first country in Asia-Pacific to enforce comprehensive AI regulation. The law designates education as a sensitive sector requiring heightened transparency and safety standards for AI systems.

We believe this law points in exactly the right direction. Mitra was built for children. Children deserve the highest standard of protection we can provide. The law asks us to document and demonstrate that protection. We are glad to do it.

This page explains what the AI Basic Act and related Korean privacy law (PIPA) require, where Mitra currently stands, and what we are actively working on as the regulatory details are finalized.

2. What the law requires, and where we stand

Requirement What it means Mitra's status
User notification Families must be told they are interacting with AI ✅ Mitra identifies herself as an AI in every interaction and in onboarding
AI-generated content labeling AI outputs must be clearly identified ✅ All responses are from Mitra, an AI system. This is never ambiguous.
Human oversight A mechanism must exist for human intervention ✅ Safety Classifier screens every message. Escalation path to human review is active.
Training data transparency Key criteria and data overview must be available ✅ See Section 4. Full corpus documentation available.
User protection plan A documented plan for protecting users must exist ✅ See Section 5. Child safety architecture documented.
Domestic representative Foreign companies must designate a Korean representative ✅ Mitra AI is registered as a Korean business entity (Reg. 332-06-03605, Gangnam-gu, Seoul)
Fundamental rights impact assessment Voluntary assessment of potential rights impacts 🔄 In progress. See Section 6.

3. Mitra is an AI. She says so.

Mitra is an artificial intelligence system. She is not a human tutor pretending to be an AI. She is not a human at all.

She is an AI tutoring companion built on large language model technology, designed specifically to support neurodivergent and neurotypical learners in Years 6-12. Every family who uses Mitra knows this from the first moment.

During KakaoTalk onboarding, every parent and student explicitly acknowledges:

"I understand that Mitra is an AI system and agree to interact with the system responsibly."

This acknowledgment is logged with a timestamp for every enrolled family.

4. How we build and verify Mitra's knowledge

Mitra's teaching is grounded in a curated corpus of peer-reviewed research on neurodivergent education, learning science, and subject-specific pedagogy. Every article in the corpus goes through a human review process before it is used to inform Mitra's teaching. We do not use unverified sources. We do not scrape the web and feed it to the model. A human being has read and approved every item in the corpus.

The corpus covers:

Mitra uses this corpus to inform her teaching strategies. She does not simply retrieve articles. She applies the pedagogical principles they contain to each student's individual context and learning style.

5. How we protect children

5.1 Safety screening, every message, every time

Every message sent to Mitra passes through a Safety Classifier before Mitra responds. The Safety Classifier checks for:

If a message triggers a safety concern, Mitra does not respond normally. She responds with care, provides appropriate resources, and the interaction is flagged for human review.

5.2 What happens in a serious situation

Mitra's conversations with students are private. We believe that trust is part of what makes Mitra effective. Students engage honestly when they know they are not being monitored. A student who feels watched will mask with Mitra the same way they mask everywhere else. That defeats the purpose.

However, safety is never private.

If a student triggers a safety concern:

  1. Mitra responds with immediate care and does not dismiss or minimize what the student has shared
  2. Mitra provides appropriate crisis resources (Korean crisis line: 1393, available 24/7)
  3. The parent or guardian is notified immediately

We recognize this approach has limitations. In complex family situations, parent notification may not always be the safest path for the child. We are actively thinking about how to handle these cases responsibly as the product matures, including future options such as student-designated trusted adults for escalation.

For now: safety triggers parent notification. Everything else stays between the student and Mitra. We believe that is the right balance for this stage of the product, and we are honest about where the edges of that policy are.

5.3 Data we collect and how we protect it

What we collect:

What we do not collect:

How long we keep it:

Who can access it:

Routine conversations are private between the student and Mitra. Parents do not receive session transcripts or conversation summaries as a matter of course. Safety events are the exception. See Section 5.2.

5.4 Parental consent and control

6. Our voluntary fundamental rights impact assessment

The AI Basic Act encourages, but does not yet require, AI providers in education to conduct a formal assessment of their system's potential impact on users' fundamental rights.

We are completing this assessment voluntarily, because we believe families deserve to see our thinking, not just our conclusions.

The assessment examines:

Right to non-discrimination. Does Mitra treat students differently based on protected characteristics? Our finding: No. Mitra does not make decisions based on gender, ethnicity, socioeconomic background, or any other protected characteristic.

Mitra does adapt her teaching approach based on learning needs that families choose to disclose, including ADHD, autism spectrum characteristics, dyslexia, and other cognitive differences. This adaptation is always in the student's favor, always based on voluntary disclosure by the family, and is the core of what makes Mitra effective for neurodivergent learners. We consider this individualized support, not differential treatment. A specialist human tutor who knows a student has ADHD teaches differently than one who does not. That is competence, not discrimination. Mitra works the same way.

Right to explanation. Can families understand how Mitra makes teaching decisions? Our finding: Yes. Mitra's teaching is grounded in a documented, human-reviewed corpus. The key criteria are: student engagement signals, prior session context, subject difficulty calibration, and neurodivergent/neurotypical learning style adaptation.

Right to privacy. Are children's data handled appropriately? Our finding: Yes. See Section 5.3. We collect the minimum data necessary, retain it for the minimum necessary period, and never share it with third parties.

Right to safety. Are adequate safeguards in place for child users? Our finding: Yes. See Section 5.1 and 5.2. Safety screening is active on every interaction. Crisis escalation path is documented and tested.

Right to human oversight. Is there meaningful human involvement in consequential decisions? Our finding: Yes. Human review is triggered by safety flags. The training corpus is human-curated. Subscription and enrollment involve human-verified processes.

Full impact assessment document available upon request to george@mitratutor.com.

7. What we are still working on

We believe in honesty about what is complete and what is in progress.

In progress:

On our watchlist:

We commit to updating this page within 30 days of any material regulatory change.

8. Questions and contact

If you have questions about how Mitra handles your child's data, or would like to request your data export or deletion, please contact:

George Neal, Founder & CEO
george@mitratutor.com
Mitra AI, Gangnam-gu, Seoul
Business Registration: 332-06-03605

We respond to all data-related requests within 5 business days.