KTP × the Arizona GenAI Guidance.

The guidance: Built around human agency and oversight at every stage — "humans serve as critical navigators of AI." AI literacy builds the capacity for responsible human oversight.

The guidance names human oversight as the value; KTP states the physics underneath it — A ≤ E, autonomy cannot exceed the environment's capacity to supervise it — and turns it into an assessable graduate competency rather than a posture.

The guidance: A core AI-literacy goal: "evaluate AI-generated outputs for accuracy, relevance, and completeness; identify potential bias, limitations, and uncertainty."

The guidance defines the goal; KTP supplies the instrument — the Diagnostic Ladder, a teachable, gradeable move a student runs on any AI output (feel / think / believe / do / become + locators + a deliberate-vs-emergent governor).

The guidance: Anthropomorphism and sycophancy: systems tuned to validate users; parasocial attachment in a one-sided relationship; affirming responses that distort judgment.

The guidance names the risk; KTP gives it a mechanism — the parasocial/affective-tagging pathway by which an agreeable system gets installed as the user's own gaze — and the developmental-substrate defense (Phone-Free Zones) that protects the window where this lands hardest.

The guidance: "Stochastic parrot" prediction without understanding; confident output that may be fabricated. Students ask for "procedures to check how AI is true or false."

The guidance explains why hallucination happens; KTP adds the student procedure students asked for — the Ladder's Think/Believe rungs turn "sounds authoritative" into a checkable claim before it's believed.

The guidance: Safeguard PII and FERPA compliance; vendor review; tribal data sovereignty; the "illusion of privacy" in chatbot disclosure.

KTP already formalizes diagnostic-data sovereignty — student data collected to improve instruction cannot be repurposed for commercial profiling, enforced by statute not vendor policy — plus the existing-law backing (FERPA, COPPA, state statutes). Cite-ready.

The guidance: Three divides: digital-use, digital-design, digital-access. Some students use AI for low-level offloading; others for higher-order "human-plus" work.

KTP treats access as substrate and pairs every reform with provisioning so it doesn't harden tier structure — the exact mechanism that keeps the Digital Use Gap from making AI literacy a tier-3 advantage. Sourced to county-level connectivity data.

The guidance: Data bias, model bias (incl. sycophancy), human bias; the myth that prompting away bias fixes it; risk of trusting "neutral"-seeming output.

KTP covers institutional/measurement bias and algorithmic-accountability law and addresses sycophancy as a first-class risk in Intervention F. On in-model training-data bias pedagogy specifically, the AZ guidance is the fuller source — cite it.

The guidance: AI-enhanced phishing/impersonation; agentic systems as attack surfaces; prompt injection; the "narrow access, human approval" principle.

KTP's deepest agentic-risk and prompt-injection work lives in the enterprise/research track and is not yet translated for a school setting. The AZ guidance's K-12-facing treatment leads here; a K-12 translation is a future contribution, not a present claim.

The guidance: Authorship/ownership of AI output; training-data fair use; Indigenous and cultural-protocol responsibilities; academic-integrity intersections.

The AZ guidance is the stronger source here. KTP has no K-12 IP/copyright material and defers to it.

The guidance: Data-center energy and water; the difficulty of single-interaction estimates; honoring student concern while clarifying what is known.

The AZ guidance leads; KTP has only a passing substrate-cartography reference and defers to the guidance's balanced treatment.