While higher exposure to AI tools and greater adult supervision significantly reduce privacy-related negative experiences among K-12 students, self-perceived understanding of data protection does not correlate with improved outcomes, revealing a critical gap between perceived and actual digital literacy.

Objective: The primary goal of this quantitative study was to investigate K-12 students' awareness of data privacy and their experiences with AI tools in educational settings. Specifically, the research aimed to examine: (1) whether students with greater exposure to AI tools exhibit higher levels of privacy awareness and concern; (2) whether higher levels of parental or teacher supervision during AI tool use are associated with lower incidence of negative experiences related to data privacy; and (3) whether students with lower self-reported understanding of data protection are more likely to report negative experiences when using AI tools and applications. The study sought to address the critical gap in empirical research on students' perspectives regarding AI in education, as most existing studies focus on technical aspects or teacher perspectives while leaving students' experiences largely unexplored.

Methods: This quantitative study employed a structured survey methodology with 125 students (ages 7-15) from William Howard Taft High School in Chicago, Illinois, USA. The sample was selected using random stratified sampling to ensure representation across key demographic segments, with balanced gender distribution (52% female, 48% male) covering both primary and early secondary school levels. The school was chosen because AI tools are regularly integrated into both curricular and extracurricular activities.

Data collection occurred via an online questionnaire administered through Google Forms. The instrument included items designed for quantitative analysis using a 5-point Likert scale (1-5), enabling examination of differences in students' attitudes across several domains: level of technology-related knowledge and skills, awareness and understanding of AI technologies, current usage of AI technologies in teaching, concerns about AI technologies, and professional development needs for AI technology.

To enhance measurement precision, multiple related survey items were aggregated into composite variables using weighted means. Student responses were weighted using the 5-point scale (1=Strongly disagree to 5=Strongly agree) across all items, then related items were aggregated to form average scores for constructing composite latent variables representing independent and dependent variables for each hypothesis.

Three hypotheses guided the investigation: H1 posited that students with greater exposure to AI tools and applications exhibit higher levels of privacy awareness and concern; H2 proposed that higher levels of parental or teacher supervision during AI tool use are associated with lower incidence of negative experiences related to data privacy; and H3 suggested that students with lower self-reported understanding of data protection are more likely to report negative experiences when using AI tools and applications.

For H1, the independent variable (exposure to AI tools) was measured by self-reported familiarity and frequency of use, while the dependent variable (privacy awareness and concern) derived from responses regarding understanding of data protection, ability to configure privacy settings, and concern about personal data collection. For H2, the independent variable (level of adult supervision) was measured by responses about using AI tools under teacher guidance at school or parent guidance at home, while the dependent variable (negative experiences) was operationalized through self-reports of unpleasant incidents resulting from data sharing. For H3, the independent variable (self-assessed understanding of data protection) was constructed from average scores on items assessing knowledge of data protection, understanding of data privacy concepts, ability to configure privacy settings, and understanding of consent for data processing.

Data analysis procedures included: assessment of normality using Kolmogorov-Smirnov and Shapiro-Wilk tests (which confirmed normal distribution, justifying parametric methods); descriptive statistics; Pearson correlation analysis; independent samples t-tests; and multiple linear regression analysis. Scale reliability was assessed using Cronbach's alpha coefficients. All statistical analyses were performed using IBM SPSS Statistics 25. The university Institutional Review Board approved the study, with voluntary participation that did not impact academic evaluations.

Key Findings: The analysis revealed several significant findings organized around the three hypotheses and supplementary regression analysis.

First, regarding scale reliability, the overall scale measuring data security and privacy in AI systems demonstrated high internal consistency (Cronbach's α = 0.908). Individual subscales showed acceptable to very good reliability: exposure to AI tools (α = 0.803), self-reported understanding of data protection (α = 0.817), privacy awareness and concern (α = 0.744), and level of adult supervision (α = 0.723), indicating that variables were well-correlated and reliably measured their intended constructs.

Second, concerning Hypothesis 1 (exposure and privacy awareness), Pearson correlation analysis revealed a weak but statistically significant positive relationship between students' exposure to AI tools and their level of privacy awareness and concern (r = 0.250, p < .01). These findings suggest that students with greater exposure to AI tools and applications tend to exhibit higher levels of privacy awareness and greater concern regarding data protection. Although the correlation is relatively low, it confirms a trend whereby increased exposure to AI technologies may be associated with heightened attention to privacy issues. Based on these findings, Hypothesis 1 was fully supported.

Third, regarding Hypothesis 2 (supervision and negative experiences), Pearson correlation analysis revealed a relatively weak but statistically significant negative correlation between the number of negative experiences related to data privacy and the level of adult supervision during AI tool use (r = -0.219, p < .05). This correlation suggests that students who use AI tools under higher levels of adult supervision are less likely to report negative experiences related to data privacy. Additionally, an independent samples t-test comparing students with lower versus higher supervision levels showed statistically significant differences in frequency of negative experiences. Students using AI tools under greater parental or teacher supervision reported lower mean levels of negative experiences (M = 1.84, SD = 1.12) compared to those with lower supervision (M = 2.43, SD = 1.09, p < .01). This finding confirms that higher levels of adult supervision may have a protective effect in reducing problems associated with data sharing and privacy violations. Based on these findings, Hypothesis 2 was fully confirmed.

Fourth, regarding Hypothesis 3 (understanding and negative experiences), Pearson correlation analysis revealed a very weak positive correlation between negative experiences related to data privacy and students' self-reported understanding of data protection (r = 0.135, p > .05). This correlation was not statistically significant, suggesting no reliable evidence to support an association between lower levels of understanding and higher likelihood of unpleasant experiences. Therefore, the data did not support the proposed hypothesis, and Hypothesis 3 was not confirmed. This counterintuitive finding indicates a critical gap between self-perceived knowledge and actual protective behaviors.

Fifth, multiple linear regression analysis was conducted to identify the most relevant predictors of negative experiences. The results revealed a statistically significant model (ANOVA, p < .01) explaining 13.1% of the variance in frequency of negative experiences with AI tools (R² = 0.131, Adjusted R² = 0.110). Three variables showed significant effects: (1) Level of exposure to AI tools showed a significant negative association with frequency of negative experiences (B = -0.362, p < .01), indicating that students more frequently exposed to AI tools report fewer negative experiences, suggesting that better familiarity with technology and understanding of data processing may reduce likelihood of unpleasant situations; (2) Self-reported understanding of data protection was positively associated with frequency of negative experiences (B = 0.214, p < .05), implying that students with greater awareness of data protection may be more capable of recognizing privacy risks and more likely to report problematic incidents rather than being genuinely more vulnerable; and (3) Level of adult supervision demonstrated a statistically significant negative effect on negative experiences (B = -0.269, p < .01), confirming that students using AI tools under parental or teacher supervision are less likely to encounter privacy-related issues. Based on these findings, both exposure to AI tools and adult supervision contribute to reduction in students' negative experiences, while higher privacy awareness may be associated with increased reporting rather than increased vulnerability.

Implications: The findings have significant implications for educational practice, policy, and research across multiple dimensions.

For educational practice and curriculum development, the study demonstrates urgent need for systematic integration of digital privacy and AI ethics education into K-12 curricula. The confirmation that self-perceived understanding does not correlate with reduced negative experiences reveals that current approaches to digital literacy education are insufficient. Schools must move beyond basic technical skills instruction to develop comprehensive modules addressing data protection, algorithmic bias, digital footprints, students' rights in technology-mediated settings, and critical evaluation of AI systems. These modules should be developmentally appropriate, interdisciplinary (involving pedagogy, technology ethics, information science, and education law experts), and focused on building actual competencies rather than superficial awareness.

For teacher preparation and professional development, the findings underscore critical need for mandatory professional development programs equipping teachers with knowledge and skills necessary for working in AI-supported environments. Currently, most U.S. teachers lack formal training in data privacy and cannot distinguish between platforms complying with FERPA/COPPA regulations and those that do not. Training must focus on: identifying risks related to data collection, algorithmic processing, and platform-based surveillance; configuring privacy settings and evaluating secure digital tools; implementing informed consent procedures with students and parents; creating inclusive, ethically grounded, and privacy-conscious learning environments; and understanding their role in mediating students' AI experiences. The study's finding that adult supervision significantly reduces negative experiences highlights teachers' protective role, but this can only be effective if educators are properly trained.

For student protection and supervision, the strong protective effect of adult supervision (reducing negative experiences) demonstrates that simply providing students with AI tools without adequate oversight is insufficient and potentially harmful. Schools must establish clear policies and practices ensuring that students, especially younger ones, use AI tools under appropriate adult guidance. However, as noted in literature, parental mediation strategies often lack coherence, and teachers may be underprepared to manage privacy risks effectively. Therefore, supervision must be informed, structured, and supported by institutional training and clear guidelines rather than left to individual adults' discretion or intuition.

For assessment and evaluation, the study's rejection of H3 reveals critical need for standardized, certified assessment tools for digital literacy that focus specifically on privacy and data ethics. Current reliance on self-reported understanding is clearly inadequate, as students' perceptions of their knowledge do not correlate with actual protective outcomes. Assessments must: measure actual competencies rather than perceived knowledge; evaluate students' critical understanding of data flows, ability to configure privacy settings, and capacity to assess risks; go beyond technical proficiency to assess ethical reasoning and decision-making in digital contexts; and provide valid, reliable measures that can inform targeted interventions for students who need additional support.

For institutional policy and infrastructure, the findings highlight systemic gaps in U.S. K-12 education regarding AI governance and data protection. Despite regulations like FERPA, CIPA, and COPPA, significant gaps remain, with vague definitions of educational necessity and minimal enforcement often leading to heightened surveillance without meaningful transparency. Schools frequently operate without clear policies or sufficient staff training regarding data privacy and ethical AI use. Nearly 90% of educational technology leaders are responsible for student data privacy, yet this role is formally recognized in only a minority of cases, with 75% reporting it's not part of official job descriptions and 17% receiving no formal training. Addressing these gaps requires: clear institutional policies governing AI tool selection, implementation, and oversight; designated, trained personnel responsible for student data privacy with formal job recognition and adequate resources; transparent data practices with enforceable accountability mechanisms; and coordination among administrators, teachers, IT staff, and families to ensure consistent, informed approaches to AI use.

For addressing systemic inequities, the study raises concerns about potential exacerbation of existing educational inequalities through AI integration. The digital divide, made evident during COVID-19, continues to affect students' access to technology, connectivity, and technical support. Without targeted policies and investments ensuring equitable access, AI tools risk widening rather than narrowing achievement gaps, leaving disadvantaged students without equal resources and potentially subject to greater surveillance without corresponding protections or benefits. Ensuring equity requires: strategic investments in infrastructure for underserved schools and communities; professional development opportunities accessible to all educators regardless of school resources; consideration of how different student populations (by race, ethnicity, socioeconomic status, disability status, language background) are affected by AI tools; and monitoring systems to identify and address disparities in AI-related experiences and outcomes.

For understanding datafication and surveillance, the study's context within U.S. schools' widespread adoption of commercial surveillance tools (Gaggle, GoGuardian) and assessment platforms (iReady) illustrates how AI integration contributes to datafication of learning environments, where students' behavioral and social interactions are transformed into measurable data points. Students are increasingly represented through algorithmic models creating "digital twins" that substitute their real identities, profiles that are not merely descriptive but shape how students are taught, assessed, and treated institutionally. This process occurs largely without students' awareness or meaningful consent, raising fundamental questions about informed consent, student autonomy, and ethical accountability that current regulatory frameworks inadequately address.

Limitations: Several limitations should be acknowledged. First, the study was conducted at a single public school (William Howard Taft High School in Chicago), which may limit generalizability to other institutional contexts, geographic regions, socioeconomic settings, or educational systems. The specific characteristics of this school—including its AI integration practices, student demographics, and institutional culture—may not be representative of K-12 schools more broadly. Second, the sample size of 125 students, while adequate for statistical analysis, is relatively modest and may limit ability to detect smaller effects or conduct more complex subgroup analyses. The age range (7-15 years) is broad, spanning developmental stages from early childhood through early adolescence, yet the study did not systematically examine how relationships between variables might differ across age groups or developmental levels. Third, the study employed a cross-sectional design using post-intervention survey data only, without pre-intervention baseline measures or longitudinal follow-up. This limits ability to establish causal relationships or track how students' awareness, experiences, and behaviors evolve over time with continued AI exposure or changing educational practices. Fourth, all key variables relied on self-reported data, which are subject to social desirability bias, recall bias, and limitations in students' ability to accurately assess their own knowledge, behaviors, and experiences. Young children in particular may have difficulty understanding abstract concepts like "data protection" or accurately reporting their experiences. The study's finding that self-reported understanding does not correlate with outcomes highlights this measurement challenge. Fifth, the study did not include objective measures of students' actual digital competencies, protective behaviors, or outcomes beyond self-reported negative experiences. Without behavioral observations, performance assessments, or actual incident data, it's difficult to validate students' self-reports or fully understand the mechanisms linking exposure, supervision, and outcomes. Sixth, the relatively modest explained variance in the regression model (R² = 0.131) indicates that the measured variables account for only 13.1% of variation in negative experiences, suggesting that many other unmeasured factors likely influence students' privacy-related outcomes. These might include specific types of AI tools used, nature and quality of supervision, individual differences in cognitive development or digital skills, family socioeconomic status, or broader school culture around technology. Finally, the study did not systematically examine potential moderating variables (such as age, gender, prior technology experience, socioeconomic status) that might influence how exposure, supervision, and understanding relate to outcomes for different student subgroups.

Future Directions: The study points to several important directions for future research. First, longitudinal research is essential to track how students' privacy awareness, understanding, and experiences evolve over time with continued exposure to AI tools and participation in digital literacy education. Such studies could examine: developmental trajectories of privacy understanding from early childhood through adolescence; long-term effects of early AI exposure and supervision on later digital competencies and behaviors; effectiveness of specific educational interventions in building sustainable privacy skills; and how relationships between exposure, supervision, understanding, and experiences change as students mature and gain experience. Second, objective measurement development is crucial given the study's finding that self-reported understanding does not predict outcomes. Future research should develop and validate: performance-based assessments of actual privacy knowledge and skills (rather than self-perceptions); behavioral measures of protective practices when using AI tools; observational protocols for assessing quality of adult supervision and mediation; and incident tracking systems that capture actual privacy violations or problematic experiences rather than relying solely on self-reports. Third, intervention research is needed to test specific approaches to improving student outcomes, including: evaluation of structured curriculum modules on data privacy and AI ethics using experimental or quasi-experimental designs; comparison of different pedagogical approaches (e.g., direct instruction vs. experiential learning vs. project-based approaches) for building privacy competencies; assessment of teacher professional development programs' effectiveness in enabling informed supervision; and testing of technology-mediated interventions (e.g., just-in-time privacy prompts, interactive tutorials) embedded in AI tools themselves. Fourth, examination of moderators and mechanisms should explore: how relationships between exposure, supervision, understanding, and outcomes differ across age groups, developmental stages, or grade levels; whether effects vary by student characteristics such as gender, socioeconomic status, prior technology experience, or cognitive abilities; what specific aspects of supervision are most protective (e.g., co-use vs. monitoring vs. active instruction vs. discussion); and how different types of AI tools (e.g., tutoring systems vs. assessment platforms vs. search tools) present different risks and benefits. Fifth, comparative and contextual research could examine: how findings generalize across different schools, districts, states, or countries with varying technology integration practices and regulatory environments; how institutional factors (school policies, leadership priorities, available resources, professional development opportunities) influence student experiences; and how cultural contexts shape students', parents', and teachers' approaches to privacy and AI use. Sixth, expanded scope research should investigate: parents' roles, knowledge, practices, and challenges in supervising children's AI use at home; peer influences on students' privacy attitudes and behaviors; students' own agency and decision-making processes when encountering privacy risks; and broader ecological factors (community norms, media influences, technology company practices) affecting student privacy. Finally, policy and practice translation research should examine: how research findings can be effectively translated into practical guidance for schools, teachers, and families; what policies and infrastructures are most effective in supporting ethical, safe AI integration; how to balance innovation and educational benefits of AI with privacy protection and student rights; and how to ensure equitable access to both AI learning opportunities and privacy protections across diverse student populations.

Title and Authors: "Between Exposure and Protection: Data Privacy Awareness and AI Tool Use Among K-12 Students" by Milena Škobo (Sinergija University, Bosnia & Herzegovina) and Milena Šović (University Business Academy in Novi Sad, Serbia).

Published on: 2025 (based on the DOI and journal volume/issue information).

Published by: The New Educational Review (DOI: 10.15804/tner.2025.81.3.02), suggesting publication in Volume 81, Issue 3 of the journal.