When Learning Signals Become Safety Signals: A Bounded-Confidentiality Framework for Educational AI Agents

Hanjing Shi, Dominic DiFranzo
Proceedings of the Impactful and Responsible AI Systems for Education Workshop, PMLR 339:123-128, 2026.

Abstract

Educational AI agents are usually introduced as tutors, homework helpers, or study companions. As children use them regularly, learning help can become a disclosure site: a student may begin with a math problem and then describe bullying, fear of adult reaction, or school avoidance. This position paper asks how child-facing educational agents should preserve student trust while deciding what stays private, what can be shared under student control, and what must be routed to accountable adults. Full parental visibility would turn help-seeking into surveillance; full secrecy would leave serious harm unsupported. We argue for bounded confidentiality as a communication-governance framework: ordinary tutoring remains private unless student choice or credible danger creates a reason to involve adults. The paper specifies a response ladder for minimal records, student-reviewed communication, care-oriented adult prompts, and escalation to locally accountable humans.

Cite this Paper

BibTeX
@InProceedings{pmlr-v339-shi26a, title = {When Learning Signals Become Safety Signals: A Bounded-Confidentiality Framework for Educational AI Agents}, author = {Shi, Hanjing and DiFranzo, Dominic}, booktitle = {Proceedings of the Impactful and Responsible AI Systems for Education Workshop}, pages = {123--128}, year = {2026}, editor = {Basu Mallick, Debshila and Woodhead, Simon and Wang, Zichao and Ananda, Muktha and Burstein, Jill and Murphy, April}, volume = {339}, series = {Proceedings of Machine Learning Research}, month = {28 Jun}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v339/main/assets/shi26a/shi26a.pdf}, url = {https://proceedings.mlr.press/v339/shi26a.html}, abstract = {Educational AI agents are usually introduced as tutors, homework helpers, or study companions. As children use them regularly, learning help can become a disclosure site: a student may begin with a math problem and then describe bullying, fear of adult reaction, or school avoidance. This position paper asks how child-facing educational agents should preserve student trust while deciding what stays private, what can be shared under student control, and what must be routed to accountable adults. Full parental visibility would turn help-seeking into surveillance; full secrecy would leave serious harm unsupported. We argue for bounded confidentiality as a communication-governance framework: ordinary tutoring remains private unless student choice or credible danger creates a reason to involve adults. The paper specifies a response ladder for minimal records, student-reviewed communication, care-oriented adult prompts, and escalation to locally accountable humans.} }
Endnote
%0 Conference Paper %T When Learning Signals Become Safety Signals: A Bounded-Confidentiality Framework for Educational AI Agents %A Hanjing Shi %A Dominic DiFranzo %B Proceedings of the Impactful and Responsible AI Systems for Education Workshop %C Proceedings of Machine Learning Research %D 2026 %E Debshila Basu Mallick %E Simon Woodhead %E Zichao Wang %E Muktha Ananda %E Jill Burstein %E April Murphy %F pmlr-v339-shi26a %I PMLR %P 123--128 %U https://proceedings.mlr.press/v339/shi26a.html %V 339 %X Educational AI agents are usually introduced as tutors, homework helpers, or study companions. As children use them regularly, learning help can become a disclosure site: a student may begin with a math problem and then describe bullying, fear of adult reaction, or school avoidance. This position paper asks how child-facing educational agents should preserve student trust while deciding what stays private, what can be shared under student control, and what must be routed to accountable adults. Full parental visibility would turn help-seeking into surveillance; full secrecy would leave serious harm unsupported. We argue for bounded confidentiality as a communication-governance framework: ordinary tutoring remains private unless student choice or credible danger creates a reason to involve adults. The paper specifies a response ladder for minimal records, student-reviewed communication, care-oriented adult prompts, and escalation to locally accountable humans.
APA
Shi, H. & DiFranzo, D.. (2026). When Learning Signals Become Safety Signals: A Bounded-Confidentiality Framework for Educational AI Agents. Proceedings of the Impactful and Responsible AI Systems for Education Workshop, in Proceedings of Machine Learning Research 339:123-128 Available from https://proceedings.mlr.press/v339/shi26a.html.

Related Material