Responsible AI in K–12 Education: A Policy Review

• In Event: Innovating K–12 Teaching and Learning with Generative AI: Insights From IES-funded National R&D Centers

Abstract

Objective: With the integration of artificial intelligence (AI) into K–12 education, ethical concerns such as algorithmic bias, data misuse, and privacy breaches disproportionately affect vulnerable student populations (Wang et al., 2024), posing risks to long-term human well-being and development (Berson et al., 2025). In response, growing policy efforts advocate for more ethical and responsible AI use in education. However, most documents focus on high-level principles, with little clarity on stakeholder responsibilities. To address this gap, we reviewed policies to develop a stakeholder-centered framework for actionable ethical guidance.

Perspectives/Methods: Using human-in-the-loop semi-automated thematic analysis (Wiebe et al., 2025), we analyze 24 AI-related education policy documents (2018–2024), sourced from national governments, international organizations, and educational research institutions. Taking the five themes of Fu and Weng’s (2024) responsible human-centered AI framework (HCAI) as a starting point, we investigate two research questions: (RQ 1) What key values and principles of responsible AI in K–12 education are reflected in policy documents? (RQ 2) How are stakeholders' responsibilities defined, and how do they interact bidirectionally with ethical principles in policy discourse?

Results: Our findings highlight three key patterns. First, while over half of the policies focus specifically on AI in education, with most documents published in the past two years, the ethical priorities remain uneven. “Fairness and Equity” and “Non-maleficence and Beneficence” received more emphasis, while “Transparency and Intelligibility”, “Privacy and Security” received less attention, and “Agency and Autonomy” were inconsistently addressed. These patterns show that current efforts lean more toward harm prevention than promoting positive outcomes, warranting greater attention moving forward. To guide future development, we extend the framework by identifying five second-level themes under each theme to clarify directions for future policy design.
Second, we find that across the reviewed policies, engagement with key stakeholders, particularly regarding their roles and responsibilities, remains insufficient, potentially limiting the policies’ operationalizability. For example, designers and developers are expected to ensure robust, explainable AI, but receive little guidance on applying culturally responsive bias mitigation. Policymakers’ roles often emphasize governance but overlook digital literacy and sustainable practices. Safeguarding measures for students prioritize privacy while neglecting emotional well-being and critical literacy. Parental roles are narrowly framed around consent and data protection, with minimal support for capacity-building or reciprocal communication. These asymmetries suggest that stronger bidirectional engagement is needed between stakeholders and ethical principles.
Finally, we identify several cross-cutting themes (e.g., community capacity building, interdisciplinary collaboration, and context-responsive infrastructure). These themes go beyond individual categories and offer potential pathways to model dynamic relationships between stakeholders and ethical principles.

Significance: Overall, this study contributes a conceptual and operationalizable framework for guiding future AI policy in education. By centering stakeholder roles and responsibilities, we suggest the need for both top-down governance and bottom-up engagement to ensure multistakeholder participation and human-centered implementation of responsible AI in K–12 education, aiming to foster more ethical, inclusive, and sustainable relationships between technology and humans.

Authors

• Zhuoyun Cai, University at Buffalo - SUNY

• Zeyu Tang, Stanford University

• X. Christine Wang, University at Buffalo - SUNY

• Sanmi Koyejo, Stanford University

• Ari Hock, University at Buffalo - SUNY

• Angelina Wang, Stanford University

• Kristen Smigielski, University at Buffalo - SUNY / CELaRAI

• Christopher Hoadley, University at Buffalo - SUNY