AI Tutoring Requires Institutional Responsibility
Full article published on VKTR — The Business of Enterprise AI
The cognitive risks of AI tools have been documented primarily in adult populations. What has received less attention is the categorical difference between what those risks mean for a developed mind and what they mean for a mind still under construction.
In a new co-authored analysis with Ryan Purdy published through VKTR, I argue that AI in education represents the third instance of a pattern that has repeated across every major category of preventable cognitive harm in the last century. Lead damaged the physical structures of the developing brain. Social media disrupted the emotional and attentional environment children developed within. AI operates on a different vector entirely: it integrates into the cognitive process itself, during the developmental window when that process is being calibrated.
The adult research has established two critical findings. Users who co-write with opinionated language models experience measurable shifts in their own stated beliefs that persist after the session ends (Jakesch et al., 2023). Users consistently accept model outputs without verification, eroding the internal calibration that distinguishes adequate answers from correct ones (Sourati, Ziabari, & Dehghani, 2026). For adults, this represents a drift from an existing cognitive position. For children, who have no prior position to drift from, the same mechanism constitutes formation. The model's distribution becomes the child's standard, not a deviation from it.
The article traces the foreseeability threshold for this category of harm and finds it has already been crossed. The EU AI Act, Colorado's AI governance law, and federal guidance from the U.S. Department of Education all signal that cognitive impact assessment in educational AI is moving from recommendation to requirement. Most school districts currently have no governance process that evaluates cognitive impact before tool authorization. Data privacy has a designated reviewer. Accessibility has a designated reviewer. Cognitive impact has no one.
The full analysis, including the governance gap, the seven minimum assessment questions drawn from impact assessment architecture, and the case that institutional responsibility precedes legal obligation, is available on VKTR.
For a technical breakdown of the architectural triggers for these systems, download our Full Working Paper.
Related Research: The Era of Cognitive Capture — ResearchGate
Author: Timothy Cook, Director, The Cognitive Privacy Project Securiti Certified in AI Security and Governance
© The Cognitive Privacy Project, 2026. CC BY-NC-ND 4.0
