Systemization of Knowledge (SoK): Human-Centered AI Safety for Youth

Pratyasha Saha; Yaman Yu; Yang Wang

cs.HCOct 6, 2026

Systemization of Knowledge (SoK): Human-Centered AI Safety for Youth

Authors: Pratyasha Saha, Yaman Yu, Yang Wang

Organizations: University of Illinois Urbana-Champaign Champaign, IL, United States

Abstract

While HCI increasingly examines AI-safety for youth, the literature lacks a comprehensive view of what risks have been identified, how they are addressed, and whether proposed protections work in-practice. We systematically reviewed 100 empirical HCI studies involving children and youth interacting with or exposed to AI across schools, homes, care settings, and public services. Using the YAIR taxonomy for risks and the MIT Mitigation Taxonomy for countermeasures, we map which risks have been identified, whether each risk is addressed by countermeasure(s), and whether each countermeasure for that risk is implemented and even evaluated. The risk-countermeasure mapping shows that most risks are matched only with proposed/ideated countermeasures; few countermeasures have been implemented, and fewer still evaluated; and existing evaluations often measure technical performance rather than protection from harm. We identify where coverage is absent, where safeguards remain untested, and propose concrete directions for HCI research to strengthen youth AI-safety.

Figures & tables

Figure 1. PRISMA Flow Diagram: Search, Screening, and Selection Process PRISMA Flow Diagram: Search, Screening, and Selection Process

Category	Keywords
Youth	“Youth,” “young,” “teen,” “adolesce,” “K-12,” “child,” “student”
AI Systems	“AI,” “GenAI,” “Generative AI,” “LLM,” “Artificial intelligence,” “chatbot,” “emotion AI,” “AI companion,” “persona,” “ChatGPT,” “conversational AI”
AI Safety and Risks	“safety,” “risks,” “harms”
Search String: (“Youth” OR young OR teen* OR adolesce* OR “K-12” OR child* OR student*) AND (AI OR GenAI OR “Generative AI” OR LLM OR “Artificial intelligence” OR chatbot OR “emotion AI” OR “AI companion” OR persona OR ChatGPT OR “conversational AI”) AND (safety OR risks OR harms)

Table 1. Keyword categories and search string used for the systematic review.

Venue	Identified	Screened	Included
CHI	209	209	71
IEEE S&P	394	379	1
CSCW / PACM HCI	64	64	10
FAccT	35	35	10
DIS	19	19	4
AIES	14	14	2

Table 2. Records identified, screened, and included per venue.

Criterion	Included studies
Publication type	Full-length research articles; excluding posters, extended abstracts, workshop papers, and book chapters.
Evidence	Empirical research.
Language	English.
Technology	AI systems.
Population	People under 18, addressed directly, through relevant stakeholder accounts, or through empirical analysis of systems or data affecting them.
Scope	Risks or countermeasures relevant to youth interaction with or exposure to AI.

Table 3. Study eligibility criteria.

Field	Recorded information
Study	Paper ID, citation, year, venue, technology, method, sample, population
Risk	Reported risk; YAIR category and numbered risk where applicable
Countermeasure	Reported countermeasure; MIT domain and subcategory
Evidence stage	Proposed; designed or implemented; evaluated
Claim provenance	Observation or evaluation; participant concern; author-anticipated; attributed to prior literature
Population relationship	First-person; boundary-spanning; stakeholder account; audit

Table 4. Data recorded for each included paper.

Figure 2. Taxonomy of AI-related risks identified for youth and the corresponding countermeasures proposed in the literature. (Part 1) A taxonomy showing categories and subcategories of AI-related risks for older adults alongside mitigation categories, mitigation subcategories, proposed countermeasures, and the papers supporting each countermeasure.

Figure 3. Taxonomy of AI-related risks identified for youth and the corresponding countermeasures proposed in the literature. (Part 2) A taxonomy showing categories and subcategories of AI-related risks for older adults alongside mitigation categories, mitigation subcategories, proposed countermeasures, and the papers supporting each countermeasure.

Stage	Measures (of 52)	Links (of 80)	Risks, by highest stage reached (of 36)
Proposed	24 (46%)	40 (50%)	11 (31%)
Designed	1 (2%)	1 (1%)	1 (3%)
Implemented	14 (27%)	25 (31%)	9 (25%)
Evaluated	13 (25%)	14 (18%)	10 (28%)
None	–	–	5 (14%)

Table 5. Maturity of countermeasures across measures, observed risk–countermeasure links, and risks.

Appendix figures & tables1 asset

Supplementary material from the paper’s appendix.

Appendix

Type	Category/Domain	Working definition
Risk	Behavioral and Social Developmental	Risks that interfere with young people’s learning, reasoning, social interaction, family relationships, or development of appropriate social boundaries.
Risk	Mental Wellbeing	Risks to young people’s emotional wellbeing, including overreliance, emotional dependency, distress, and inadequate support during vulnerable situations.
Risk	Toxicity	Exposure to harmful, inappropriate, threatening, sexual, violent, or otherwise age-inappropriate content.
Risk	Misuse and Exploitation	Risks arising when AI is used to harm, manipulate, deceive, coerce, or exploit youth, or when protections are circumvented.
Risk	Bias/Discrimination	Risks involving stereotyping, distorted representation, unequal access or treatment, and normative judgments that disadvantage particular youth or groups.
Risk	Privacy	Risks involving inappropriate collection, processing, disclosure, or representation of information about young people.

Appendix

Table 6. Working definitions of the broad risk categories and countermeasure domains used in this review.

Explore similar work

Beyond "I Can't Help With That": How Child Safety Experts Evaluate AI Chatbot Safety
Aug 8, 2026Hannah Cha, Neha Shukla, Solon Barocas +3Artificial Intelligence Safety Psychology
Microsoft Research · Duke University · Abridge AI Inc. +1
Position: Preventing AI-Generated CSAM Necessitates New Approaches to AI Safety
Jun 9, 2026Neil Kale, Rebecca Portnoff, Pratiksha Thaker +5Artificial Intelligence Safety Child Sexual Abuse Material
1Carnegie Mellon University · 2Thorn
CAREBench: A Child-Safety Risk Benchmark for Language Models
Jun 29, 2026Kaavya Krishna-Kumar, Elaine Lau, Vaughn Robinson +6Artificial Intelligence Safety Frontier Models
Handshake AI · University of California, Los Angeles · McGill University

Aug 8, 2026, cs.CYJ/K move · Enter open · S save

Beyond "I Can't Help With That": How Child Safety Experts Evaluate AI Chatbot Safety

Hannah Cha, Neha Shukla, Solon Barocas, Alexandra Chouldechova+2

Microsoft Research · Duke University · Abridge AI Inc.+1

Artificial Intelligence Safety Psychology Experts

Open paper AskPDF arXiv

Youth increasingly turn to AI chatbots for social and emotional support, raising concerns about how these systems respond, especially in high-stakes situations. However, existing child safety evaluations of AI lack grounding in real-world harms that youth experience, rely on unvalidated assumptions about what counts as an appropriate output (e.g., refusal), and typically focus on detecting adversarial prompts or surface-level harms in outputs only. Thus, these evaluations can fail to detect responses that pose harm to youth in practice. To better understand the limitations of current evaluation practices, we conducted interviews with 19 practitioners working directly with youth in vulnerable situations, including social workers, therapists, and psychologists, asking them to reflect on chatbots' responses to risky situations commonly faced by youth, as established in prior empirical work. Practitioners identified chatbot behaviors likely to cause harm as well as those that could meaningfully support youth in difficult moments, discussed the role that chatbots should (and should not) play in these interactions, and offered concrete recommendations for improving chatbot responses. Based on these findings, we provide recommendations for AI child safety evaluation and infrastructure, and highlight the need for incorporating practitioners' perspectives into safety work.

Abstract

Explore similar work

Beyond "I Can't Help With That": How Child Safety Experts Evaluate AI Chatbot Safety

Position: Preventing AI-Generated CSAM Necessitates New Approaches to AI Safety

CAREBench: A Child-Safety Risk Benchmark for Language Models