From AI Moratoriums to Developmentally Admissible AI
Generative artificial intelligence creates an educational governance problem when a technology category is treated as though every capability, learner context, objective, and realized model behavior were equivalent. This paper develops Developmentally Admissible AI (DAAI) as a transition-governance framework for student-facing AI.
Central proposition
Technology identity alone is not a sufficient governance object for every educational AI interaction; a proposed state transition can represent materially relevant distinctions that technology identity cannot.
The framework asks whether a proposed AI-mediated transition is admissible for a represented learner state, educational objective, context, governing constraints, evidence, and uncertainty. It does not presume that fine-grained governance is always preferable to coarse restriction.
Why transition governance matters
A system that writes an assessed essay, asks Socratic questions, translates instructions, reads text aloud, identifies a misconception, or generates a practice problem may rely on the same underlying model while creating materially different educational interventions. DAAI therefore separates model identity from requested capability, educational objective, context, evidence, and realized behavior.
The framework gives special treatment to cognitive agency and productive struggle. It distinguishes assistance that preserves the learner's participation in a target cognitive operation from assistance that materially performs that operation on the learner's behalf. That distinction is analytical rather than a validated psychometric measure, and substitution by itself does not establish benefit or harm.
Governed dispositions
A DAAI resolver can preserve at least five distinct outcomes: ALLOW, DENY, REVIEW, CONSTRAIN, and FAIL_CLOSED. A denial or review is still a state-producing governed resolution because provenance and system state change even when the requested educational effect does not occur.
CONSTRAIN means the original requested effect is not authorized while a lower-capability successor offer may separately be admissible.
Development, accessibility, and uncertainty
Age can be relevant without being a sufficient decision function. Developmental state may also involve demonstrated mastery, language proficiency, accessibility needs, AI literacy, prior attempts, and independence in the target task. The paper rejects turning this richer state into a license for unnecessary surveillance or durable learner labeling.
Accessibility support is treated separately from cognitive substitution. A translation, read-aloud, speech-to-text, or alternative explanation can preserve equivalent access, but an accommodation should not silently perform the very cognitive operation being assessed unless that substitution is itself authorized.
When material evidence is missing or authority conflicts remain unresolved, the framework preserves uncertainty through review, constraint, or fail-closed outcomes rather than converting missing evidence into false certainty.
Bounded NYC 2026 case study
The paper uses New York City's 2026 student-facing generative-AI moratorium as a bounded governance case study. A moratorium can be rational as a temporary uncertainty-containment state while institutions improve evaluation, teacher training, privacy controls, accessibility policy, AI literacy, learning-outcome measurement, provenance, and incident handling.
The longer-term question is not simply whether AI should be allowed. It is which AI-mediated transitions are admissible under which conditions, including the possibility that a coarse restriction remains preferable when fine-grained governance would require disproportionate surveillance, stale profiling, or operational complexity.
Evidence boundary
This is a conceptual and research framework. Publication does not establish empirical validity of DAAI, policy superiority, child-safety certification, real-time measurement of cognitive agency, or universal validity of the target-operation formalism.
Independent conceptual review found the framework internally coherent with reservations that were incorporated into the research manuscript. The reservations narrow the claims and preserve important uncertainties.
Sources and supplementary materials
- New York City Mayor's Office — 2026 student-facing generative-AI policy announcement
- Bastani et al. (2025), PNAS — Generative AI without guardrails can harm learning
- Kestin et al. (2025), Scientific Reports — AI tutoring RCT
- Oreopoulos & Low (2026), NBER — Khanmigo two-year school experiment
- Richmond & Taylor (2025), Nature Reviews Psychology — cognitive offloading
- Misiejuk et al. (2026) — longitudinal student–AI prompting strategies
- StegVerse Papers — public research index
The complete bibliography and formal research package remain part of the governed publication record. This Site page is a public presentation surface, not an independent validation authority.