A Child-Safe Multilayer Conversational AI Safety Architecture (CMCASA) and Conversational Developmental Digital Biomarkers (CDDB) for Preschool Children's Developmental Assessment: A Conceptual Framework

doi:10.15683/kosdi.2026.9.30.1063

All Issue

2026 Vol.22, Issue 3 Preview Page Next Page

Original Article

A Child-Safe Multilayer Conversational AI Safety Architecture (CMCASA) and Conversational Developmental Digital Biomarkers (CDDB) for Preschool Children's Developmental Assessment: A Conceptual Framework 아동의 발달 평가를 위한 대규모 언어 모델 기반 아동 안전 다층 대화형 인공지능 아키텍처: 검토 및 개념적 프레임워크: 이정록¹ㆍ박건철²ㆍ신원선³ㆍ정상⁴*
Jeong Rok Lee¹, Keon Chul Park², Won Sun Shin³, Sang Jeong⁴*; ¹CTO, AI Leader Inc, Cheonan, Republic of Korea
²Assistant Professor, School of Business Administration, Tech University of Korea, Siheung, Republic of Korea
³CEO, Vision21Tech Inc, Daejeon, Republic of Korea
⁴Research Professor, Office of Research Affairs, Chung-Ang University, Seoul, Republic of Korea

30 September 2026. pp. 1063-1071

PDF

Abstract

Purpose: This study aims to address safety, developmental appropriateness, privacy, and human oversight issues in LLM-based conversational systems for preschool children, proposing an integrated framework that combines child safety with developmental observation. Method: Through a literature review on child-LLM interaction risks, developmental psychology, and international regulations (UNICEF, EU AI Act, COPPA), the study proposes CMCASA—a multilayer safety architecture consisting of rule-based safety, context-aware LLM safety, and multidisciplinary human review layers—and conceptualizes CDDB as explainable developmental observation markers extracted from natural conversations. Result: CMCASA provides a multi-layered governance structure that addresses explicit harmful content, emotional risks, privacy violations, and developmental inappropriateness, while the CDAE and CDDB offer continuous, explainable developmental observation evidence rather than diagnostic scores from naturalistic child-AI conversations. Conclusion: CMCASA integrates safety, developmental assessment, privacy, and human oversight into a single governance structure; however, the reliability, construct validity, cultural generalizability, and long-term efficacy of CDDB require systematic empirical validation before real-world deployment.

Keywords

Conversational AI

Child Safety

Preschool Children

Developmental Assessment

Child-Computer Interaction (CCI)

연구목적: 미취학 아동 대상 LLM 기반 대화형 시스템의 안전성, 발달적합성, 프라이버시 및 인간 감독 문제를 통합적으로 검토하고, 아동 안전과 발달 관찰을 결합하는 개념적 프레임워크를 제안한다. 연구방법: 아동-LLM 상호작용 위험, 발달 심리학 및 국제 규제(UNICEF, EU AI법, COPPA) 관련 문헌을 고찰하여, 규칙기반 안전 계층, 맥락인식 LLM 안전 계층, 다학제적 인간 검토 계층으로 구성된 CMCASA를 제안하고, 자연스러운 대화에서 추출되는 발달 특징을 CDDB로 개념화한다. 연구결과: CMCASA는 명시적 유해성, 정서적 위험, 개인정보 침해 및 발달 부적합성을 다층적으로 관리하는 거버넌스 구조를 제공하며, CDAE와 CDDB는 진단 점수가 아닌 지속적이고 설명가능한 발달 관찰 증거를 제공한다. 결론: CMCASA는 안전성, 발달 평가, 프라이버시, 인간 감독을 하나의 거버넌스 구조로 통합하였으나, CDDB의 신뢰도, 구성 타당도, 문화적 일반화가능성 및 장기적 효과는 실제 적용 전 체계적인 실증 연구를 통해 검증되어야 한다.

키워드

대화형 인공지능

아동안전

미취학 아동

발달평가

아동-컴퓨터 상호작용

References

5Rights Foundation (2025). Children & AI Design Code: A Protocol for the Development and Use of AI Systems That Impact Children. London.
Arif, S., Borah, A., Mihalcea R. (2026). The Age of Curiosity Meets the Age of AI: Benchmarking Child Safety in Large Language Models. arXiv:2605.25510.
European Parliament and Council (2024). Regulation (EU) 2024/1689 Laying Down Harmonised Rules on Artificial Intelligence (Artificial Intelligence Act). Official Journal of the European Union, L series. http://data.europa.eu/eli/reg/2024/1689/oj
Federal Trade Commission (2025). Children's Online Privacy Protection Rule; Final Rule. 16 CFR Part 312, Federal Register.
Gilkerson, J., Richards, J.A., Warren, S.F., Montgomery, J.K., Greenwood, C.R. (2017). “Mapping the early language environment using all-day recordings and automated analysis.” American Journal of Speech-Language Pathology, Vol. 26, No. 2, pp. 248-265. 10.1044/2016_AJSLP-15-0169
Goldman, L., Poulin-Dubois, E. (2024) “Children's anthropomorphism of inanimate agents.” WIREs Cognitive Science, Vol. 15, e1676. 10.1002/wcs.1676
Jeong, Y.S., Kim, Y.-W., Yim, J. (2025). “A study on the development of an automatic classification system for life safety prevention service reporting images through the development of AI learning model and AI model serving server.” Journal of the Society of Disaster Information Vol. 19, No. 2, pp. 432-438.
Jiao, J., Afroogh, S., Chen, K. J., Murali, A., Atkinson, D. (2026). “LLMs and childhood safety: Identifying risks and proposing a protection framework for safe Child-LLM interaction.” arXiv:2502.11242. 10.2139/ssrn.6836268
Krishna-Kumar, K., Lau, E., Robinson, V., Caldwell, J., Issaka, S. (2026). “CAREBench: A child-safety risk benchmark for language models.” arXiv:2606.29685.
Kurian, N. (2024) “'No, Alexa, No!': Designing child-safe AI and protecting children from the risks of the 'empathy gap' in large language models.” Learning, Media and Technology, Vol. 50, No 4, pp. 621-634. 10.1080/17439884.2024.2367052
Kurian, N. (2025). “Developmentally aligned AI: A framework for translating the science of child development into AI design.” AI, Brain and Child, Vol. 1, No. 1, Article. 9 10.1007/s44436-025-00009-z
Lee, J., Cho, H., Yun, J.-W., Lee, H., Lee, J.K. (2025). “SGuard-v1: Safety Guardrail for Large Language Models.” arXiv:2511.12497.
Sagae, K. (2021). “Tracking child language development with neural network language models.” Frontiers in Psychology, Vol. 12, 674402. 10.3389/fpsyg.2021.674402
UNICEF (2025). Guidance on AI and Children. 3rd ed., UNICEF Office of Research – Innocenti, Via degli Alfani, 58, 50121 Firenze FI, Italy.
Wang, Y., Williams, R., Dilley, L., Houston, D.M. (2020). “A meta-analysis of the predictability of LENA^TM automated measures for child language development.” Developmental Review, Vol. 57, 100921. 10.1016/j.dr.2020.100921

Information

Publisher :The Korean Society of Disaster Information
Publisher(Ko) :한국재난정보학회
Journal Title :Journal of the Society of Disaster Information
Journal Title(Ko) :한국재난정보학회논문집
Volume : 22
No :3
Pages :1063-1071
DOI :https://doi.org/10.15683/kosdi.2026.9.30.1063

[1] 5Rights Foundation (2025). Children & AI Design Code: A Protocol for the Development and Use of AI Systems That Impact Children. London.

[2] Arif, S., Borah, A., Mihalcea R. (2026). The Age of Curiosity Meets the Age of AI: Benchmarking Child Safety in Large Language Models. arXiv:2605.25510.

[3] European Parliament and Council (2024). Regulation (EU) 2024/1689 Laying Down Harmonised Rules on Artificial Intelligence (Artificial Intelligence Act). Official Journal of the European Union, L series. http://data.europa.eu/eli/reg/2024/1689/oj

[4] Federal Trade Commission (2025). Children's Online Privacy Protection Rule; Final Rule. 16 CFR Part 312, Federal Register.

[5] Gilkerson, J., Richards, J.A., Warren, S.F., Montgomery, J.K., Greenwood, C.R. (2017). “Mapping the early language environment using all-day recordings and automated analysis.” American Journal of Speech-Language Pathology, Vol. 26, No. 2, pp. 248-265. 10.1044/2016_AJSLP-15-0169

[6] Goldman, L., Poulin-Dubois, E. (2024) “Children's anthropomorphism of inanimate agents.” WIREs Cognitive Science, Vol. 15, e1676. 10.1002/wcs.1676

[7] Jeong, Y.S., Kim, Y.-W., Yim, J. (2025). “A study on the development of an automatic classification system for life safety prevention service reporting images through the development of AI learning model and AI model serving server.” Journal of the Society of Disaster Information Vol. 19, No. 2, pp. 432-438.

[8] Jiao, J., Afroogh, S., Chen, K. J., Murali, A., Atkinson, D. (2026). “LLMs and childhood safety: Identifying risks and proposing a protection framework for safe Child-LLM interaction.” arXiv:2502.11242. 10.2139/ssrn.6836268

[9] Krishna-Kumar, K., Lau, E., Robinson, V., Caldwell, J., Issaka, S. (2026). “CAREBench: A child-safety risk benchmark for language models.” arXiv:2606.29685.

[10] Kurian, N. (2024) “'No, Alexa, No!': Designing child-safe AI and protecting children from the risks of the 'empathy gap' in large language models.” Learning, Media and Technology, Vol. 50, No 4, pp. 621-634. 10.1080/17439884.2024.2367052

[11] Kurian, N. (2025). “Developmentally aligned AI: A framework for translating the science of child development into AI design.” AI, Brain and Child, Vol. 1, No. 1, Article. 9 10.1007/s44436-025-00009-z

[12] Lee, J., Cho, H., Yun, J.-W., Lee, H., Lee, J.K. (2025). “SGuard-v1: Safety Guardrail for Large Language Models.” arXiv:2511.12497.

[13] Sagae, K. (2021). “Tracking child language development with neural network language models.” Frontiers in Psychology, Vol. 12, 674402. 10.3389/fpsyg.2021.674402

[14] UNICEF (2025). Guidance on AI and Children. 3rd ed., UNICEF Office of Research – Innocenti, Via degli Alfani, 58, 50121 Firenze FI, Italy.

[15] Wang, Y., Williams, R., Dilley, L., Houston, D.M. (2020). “A meta-analysis of the predictability of LENA^TM automated measures for child language development.” Developmental Review, Vol. 57, 100921. 10.1016/j.dr.2020.100921

Journal of the Society of Disaster Information ISSN:1976-2208(Print) 2671-5287(Online) 한국재난정보학회논문집

All Issue