AI Literacy as a Civic Capacity.
The test isn't whether students can use AI. It's whether they can keep themselves at the center of it.
The assessed skill needs a teachable instrument. The student Diagnostic-Ladder curriculum is the classroom form — the move a student runs on any AI output — adapted from the Diagnostic Ladder.
A tenth grader asks a chatbot whether her essay's argument holds. It tells her it's insightful and well-reasoned, and suggests three small polish edits. She feels good and turns it in. The model never evaluated her argument; it predicted the words most likely to keep her engaged. She has just been graded by something designed to flatter her, and she did not notice. She will make decisions like this for the rest of her life.
Most of what is sold as 'AI literacy' is tool-use: how to prompt, which app to pick, how to get a better output. That is real but it is the smaller half. The 2026 Arizona K-12 GenAI guidance gets the larger half right — it names 'human agency and oversight' as the overarching value and 'critical evaluation of AI outputs' as a core literacy goal. What no standard yet supplies is the instrument: the graded, teachable move a student actually runs on a machine answer. Absent an assessed instrument, 'critical evaluation' stays a value statement on a curriculum map, and what gets taught is tool-fluency, because tool-fluency is what can be demonstrated.
Students become fluent operators of systems they cannot evaluate. They can prompt, iterate, and ship — and they cannot tell sycophancy from feedback, a confident hallucination from a fact, or a steered conclusion from their own. This is the danah-boyd media-literacy trap one substrate deeper: literacy taught as tool-use inside an attention-optimized architecture does not return judgment; it produces faster, more confident consumption. The civic precondition — a population that can be told something by a machine and still decide for itself — is not being built.
What is graded becomes what is taught. The leverage point is making the GRADED skill the evaluation of AI outputs and the exercise of human oversight, not the use of the tool. Concretely: assess whether a student can run a structured evaluation on an AI answer (the Diagnostic Ladder), name where human judgment must override the machine (A≤E — autonomy cannot exceed the environment's capacity to supervise it), and defend the decision. Anchor it to skills the standards already prize — in Arizona, the State 48 Essential Skills (Critical Thinking, Ethics & Impact, Digital Fluency) and AI Literacy Goals #4 (Critical Evaluation) and #5 (Human Agency and the Human Advantage).
A cross-disciplinary AI-literacy standard whose ASSESSED outcome is evaluation-and-oversight, delivered through a portable instrument rather than a tool list. The instrument: the Diagnostic Ladder (run on any AI output — what does this want me to feel, think, believe, do, become; who benefits; deliberate or emergent?) plus a human-oversight rubric (where must a human stay in the loop, and what is the environment's capacity to catch the machine when it's wrong?). Taught from elementary on (much of it unplugged), graded from middle school up, woven across content areas rather than siloed into a computer-science elective.
Students are already AI-native — they don't need a class to use these tools, and a 'literacy' requirement is adults panicking about a thing kids already understand. Worse: any graded evaluation framework will be one more compliance hoop, and a rubric for 'questioning AI' is itself a way to install the rubric-writer's views.
Using is not evaluating — the AZ guidance's own students say they cannot tell when an AI answer is true, and that fluency without evaluation leaves them 'passive consumers more vulnerable to misinformation, manipulation, and algorithmic bias.' The native-user objection confuses comfort with the tool for judgment about the tool. The compliance-hoop risk is real and is why the instrument is governed: the Diagnostic Ladder's final rung is the deliberate-vs-emergent test, an explicit guard against the framework becoming the thing it warns about — a student is taught to point it at the instrument itself.
- Prompt-engineering curriculum vendors (literacy redefined as tool-use sells courses)
- Engagement-optimized AI-tutoring products whose retention depends on students NOT evaluating the output (the sycophancy incentive)
- AI-detector vendors (literacy reframed as enforcement, not capacity)
- Standardized-testing orthodoxy ('evaluation and oversight are unmeasurable, therefore unaccountable')
- Time-on-task hawks: instructional minutes are already overcommitted
- 'Kids already know this — they use AI more than we do.'
- 'You can't grade "questioning AI" objectively.'
- 'A framework for evaluating AI just installs the framework-writer's bias.'
- 'This belongs in one computer-science class, not across the curriculum.'
- 'Teach them to use the tools well — evaluation will follow.'
If three or more states adopt an AI-literacy standard whose assessed outcome is evaluation-and-oversight and, after five years, graduating students cannot run a structured evaluation on a novel AI output, identify where human judgment must override it, and articulate why — the intervention has failed in the form deployed. Revision required.
- Arizona Institute for Education and the Economy (AIEE), NAU — 2026 K-12 GenAI guidance names the goal (human agency, critical evaluation); this supplies the instrument. Natural collaborator.
- Digital Promise, aiEDU, AI4K12 — AI-literacy framework builders; aligned on literacy-as-judgment, not tool-use
- Common Sense Media — Mental-well-being and companion-AI research; aligned on the sycophancy/parasocial risk
- AFT and NEA — Aligned in principle; cautious about workload and another graded mandate
- Prompt-engineering course vendors; engagement-optimized tutoring products — Direct revenue threat from redefining literacy as evaluation, not use
- Computer-science-standards bodies — Tend to silo AI into a CS elective; the cross-disciplinary framing reads as encroachment
The literacy gap IS the AZ guidance's 'Digital Use Gap': some students use AI only for low-level offloading while others use it for higher-order 'human-plus' work. Evaluation-and-oversight is exactly the capacity that separates the two. Without it — and without Intervention D (substrate provisioning) — AI literacy becomes another tier-3 advantage: resourced students learn to direct and audit the machine, under-resourced students get the easy-button. PAIR WITH C, B, AND D.
Delivered as a standalone tech course requiring devices and connectivity, it excludes exactly the students most exposed to unevaluated AI. Mitigation: unplugged elementary lessons, cross-curricular delivery (no new device dependency), and provisioning (D) so the instrument reaches every student, not just the ones with broadband at home.
Small-group settings where a teacher can hear a student defend an evaluation; humanities-and-CS-fluent staff who can teach judgment about machines, not just operation of them; time on the school day for non-tool-demonstration work. All three are tier-3 features today.
State boards of education hold rulemaking authority over curriculum standards and graduation competencies under most state administrative procedure acts. AI-literacy outcomes can be anchored to existing standards (in Arizona, the State 48 Graduate Profile / Essential Skills and the AZ Academic Standards) without new statutory authority.
Staff fluent enough in how these systems work to teach evaluation (not just use); time on the school day not consumed by test prep (depends on A); small-group settings for defended evaluation (depends on D); a portable, governed instrument so the standard is teachable and assessable rather than aspirational (the Diagnostic Ladder fills this).
State board standards and graduate-profile competencies are revised on multi-year cycles; the instrument itself is versioned and can update as the technology shifts, without reopening the standard. The governor rung makes the framework self-correcting by design.
Students can run a structured evaluation on an AI output and name where a human must stay in the loop. Sycophancy, hallucination, and steering become things they can see rather than absorb.
Over a 10-15 year horizon, cohorts age into the electorate and workforce able to work alongside intelligent systems while keeping judgment human — the WEF Future of Jobs 'human-centric skills' differentiator. The vendor incentive shifts: a generation that evaluates outputs is a worse market for flattery-optimized products and a better one for genuinely useful tools.
State AI-literacy standards / graduate-profile competency revision (state board). District-level adoption through model curriculum and the classroom rubric. In Arizona specifically: alignment to the State 48 Essential Skills and the AZ AI Literacy Goals already articulated by AIEE.
None. Curriculum and graduation competencies are reserved to states under standard federalism.
2026-2030. The window is open NOW — authoritative state guidance (Arizona's 2026 GenAI guidance and peers) has named the goal and is searching for instruments. Supplying a governed, teachable instrument into that demand is higher-leverage than waiting for a panic to crest.
- Arizona Institute for Education and the Economy (AIEE), Northern Arizona University (2026). Generative Artificial Intelligence in K-12 Education: Guidance for Arizona Schools and School Systems (v26.01). azk12.ai
- World Economic Forum (2025). Future of Jobs Report 2025. World Economic Forum.
- Common Sense Media (2025). Talk, Trust, and Trade-offs: How and Why Teens Use AI Companions.
- boyd, d. (2018). You Think You Want Media Literacy… Do You?. Data & Society.
Corrections via /advocacy/contribute. Public response SLA: 14 days for substantive corrections, 30 days for elaborations. No comments section by design.
Submit a correction →State boards of education (curriculum standards and graduate-profile competencies); teacher-preparation accreditation bodies (can require AI-evaluation competency); district professional-development offices (can mandate cross-curricular delivery).
State curriculum audit finding AI literacy delivered as tool-use only; graduate-profile review finding no assessed evaluation/oversight competency; community or board request for a teachable instrument behind a 'critical evaluation' standard.
Accreditation conditional status; loss of AI-literacy or civics funding streams tied to the competency. No strong federal mechanism — enforcement rests on state-board commitment.
No national standard for what constitutes 'AI-output evaluation' as an assessed skill — this is pre-political. The Diagnostic Ladder is offered as a portable instrument precisely to give the standard something concrete to require; until adopted, enforcement depends on local commitment.
- 1 — clearly defined boundaries (what the assessed evaluation/oversight skill is, vs. tool-use)
- 7 — recognized right to organize (students/parents/teachers may propose instrument revisions)
- 3 — collective-choice arrangements (those taught the instrument have standing to refine it)
Students assessed on the instrument, the teachers delivering it, and parents — through the same participatory channel as Intervention E.
The instrument (the Diagnostic Ladder) is published openly and governor-gated; it cannot be enclosed into a proprietary 'AI literacy' product without surfacing that it has dropped the evaluation/oversight substance.
AI-output evaluation and human oversight are almost entirely unmeasured. Most existing 'AI literacy' metrics count tool access or use; the KPIs here deliberately measure the assessed capacity — can a student evaluate an output and keep judgment human — which requires new instruments. This is itself an argument for the Diagnostic Ladder as a portable, gradeable instrument.
The percentage of students assessed, via a documented instrument, on their ability to evaluate an AI output — identify sycophancy, hallucination, bias, and steering — as opposed to their ability to operate AI tools. This is the substance variable: it separates literacy-as-judgment from literacy-as-tool-use.
No national measurement exists. Most state and district 'AI literacy' frameworks as of 2026 assess tool familiarity or acceptable-use compliance, not output evaluation. Effective baseline near zero.
At least 30% of public middle and high schools assess AI-output evaluation through a documented instrument by 2030; the assessed skill, not tool-use, is the graded outcome.
State curriculum/graduate-profile audits; district assessment-instrument review; survey of AI-literacy standard adoption distinguishing evaluation from use.
State departments of education AI guidance and graduate-profile competencies; Digital Promise / aiEDU framework adoption data; AIEE (Arizona) AI Literacy Goals implementation.
The number of states whose graduate-profile or graduation competencies include an assessed 'human oversight / human-in-the-loop' outcome — the ability to name where human judgment must override an automated system and why. Operationalizes the AZ guidance's overarching value (human agency and oversight).
As of 2026, several states name human oversight as a value in AI guidance (Arizona's State 48 framing among the most developed); essentially none assess it as a graduation competency.
At least 5 states adopt an assessed human-oversight competency in their graduate profile by 2031.
State graduate-profile / Portrait-of-a-Graduate competency review; AI guidance content analysis distinguishing stated value from assessed competency.
State graduate-profile frameworks (e.g., Arizona State 48 Graduate Profile); Education Strategy Group Portrait-of-a-Graduate tracker.
The percentage of adopting schools delivering AI literacy across content areas (ELA, social studies, science, etc.) rather than confining it to a single computer-science elective. Cross-disciplinary delivery is what reaches every student and ties evaluation to disciplinary judgment; siloed delivery reaches the self-selected few.
Where AI literacy exists in 2026 it is most often a CS or ed-tech elective. Cross-curricular delivery is rare and undocumented.
At least 40% of adopting schools document cross-disciplinary AI-literacy delivery by 2031.
District curriculum-map audit; teacher-survey of where AI-evaluation instruction occurs.
ISTE / Digital Promise implementation surveys; state curriculum review.
The difference in assessed AI-output-evaluation capacity between Title I and non-Title I schools — a direct measure of the AZ guidance's 'Digital Use Gap.' The risk this KPI guards against: AI literacy becoming a tier-3 advantage where resourced students learn to direct and audit AI while under-resourced students get the easy-button.
No measurement exists; the digital-use gap is documented qualitatively (NETP 2024) but not as an evaluation-capacity differential.
Establish the Title I / non-Title I evaluation-capacity differential by 2029 and set a closing target; the gap narrows rather than widens as adoption scales.
Disaggregated F1 results by Title I status; paired with Intervention D provisioning data.
State assessment data disaggregated by Title I status; National Educational Technology Plan (NETP) 2024 digital-use-gap framing.
Where no national baseline exists, the KPI is still listed. The absence of a baseline is itself the measurement problem — and the first advocacy deliverable is establishing one. Targets marked as contingent on baseline establishment should be read as agenda items for the measurement reform intervention (A), not deferrals.
The substrates this intervention engages, organized by family. Bottleneck substrates (★) are the upstream layers that determine whether every later argument is sane or distorted.
- Assessment substrate#02
- ★Curriculum substrate#44
- Pedagogical substrate#45
- Instructional-materials substrate#46
- Standards substrate#47
- Intervention substrate#51