TLDR
Executive briefThe same course can authorise radically different risks
- Define AI literacy as demonstrated ability inside a named role, work event, authority, system and consequence, not as recall of generic concepts.
- Use eight domains spanning system recognition, task judgment, interaction, evidence, data, oversight, social consequence and operational stewardship.
- Qualify people through representative normal, ambiguous, failure, manipulation, population and escalation scenarios, including appropriate refusal.
- Keep a person-role evidence passport that grants bounded authority, names support and expires on material change without becoming an employee score.
- When people repeatedly fail, inspect interface, defaults, authority, workload and workflow before prescribing another course or blaming the user.
Two colleagues complete the same enterprise course. Both can define a large language model, recognise a fabricated answer and write an acceptable prompt. The first uses an approved assistant to improve the tone of low-consequence internal email. The second approves an agent that can change customer records and initiate refunds. Their shared orientation is useful, but it does not establish equivalent readiness. The second role requires evidence about tool authority, data boundaries, failure handling, affected-party recourse, monitoring, rollback and the sponsor's power to stop the service.
AI literacy is therefore not a single transferable skill. It is the demonstrated ability to act within a role's AI authority and consequence. The requirement changes with the work event, affected population, system behaviour, autonomy, reversibility, available evidence and decision right. A customer adviser receiving suggestions, an engineer connecting tools, a manager redesigning a workflow, a risk specialist challenging evidence and an executive authorising deployment each need a different pattern of judgment. Technical depth matters where the work demands it. It should not become a prestige proxy that lets technical specialists ignore business consequence or forces every worker to learn model construction.
Current public guidance supports context-sensitive measures rather than one universal syllabus. Article 4 of the EU AI Act requires providers and deployers within scope to take measures that support a sufficient level of AI literacy, considering knowledge, experience, education, training, context of use and the people affected.13 The European Commission's practice repository illustrates approaches but cautions that copying a listed practice does not itself demonstrate compliance.2 UK, Canadian, OECD and UNESCO materials separate technical, responsible, ethical, cognitive and occupational capabilities, while NIST links effective risk work to defined roles, context and lifecycle action.46710
This paper turns that principle into an operating architecture. The Eight-Domain AI Literacy Map defines what can matter. The Role-Consequence Literacy Matrix selects what a person actually needs for one role. Five Qualification States separate orientation, assisted practice, qualified operation, consequential stewardship and independent challenge. The Scenario Qualification Protocol observes action and refusal in realistic work. A Literacy Evidence Passport records the authority and support earned by each person-role pairing. The Literacy-to-Control Bridge prevents repeated failure from being misdiagnosed as an individual training deficit.
The boundary is deliberate. Paper 040 made role, proficiency, support and workload part of change-bearing architecture. Paper 045 allocated decision rights. Paper 109 designed human oversight that can alter an outcome. Paper 131 asks a narrower question: what evidence qualifies this person to exercise or challenge a specific AI authority? It does not build the broader workforce capability model reserved for Paper 132 or the learning architecture reserved for Paper 138. Training completion is an input. Qualification is observed performance under the conditions the role will face.
Eight-Domain AI Literacy MapReplace the prompt certificate with a map of situated judgment
The Eight-Domain AI Literacy Map is this article's synthesis. It begins with recognition and system boundaries: noticing where AI is present, identifying the configured service, intended purpose, dependencies and limits. A worker cannot make a defensible choice if an embedded recommendation looks like an ordinary database field or if the name of a consumer product hides the enterprise configuration actually in use.
Task and tool judgment asks whether AI is suitable for the work at all, what remains human, which approved mode fits and when a non-AI route is better. Interaction and specification covers framing the task, providing lawful context, stating constraints, steering iterations and preserving important assumptions. Prompt technique sits here, as one practice inside a larger decision about whether and how the tool belongs in the task.
Output and evidence judgment means testing claims, inspecting sources, detecting uncertainty and bias, comparing alternatives, recording checks and refusing unsupported work. Fluency can make a weak claim feel settled, so the role must know what evidence would change the decision. Data, privacy and security covers information classification, rights, provenance, approved channels, leakage, manipulation and protected data. These duties belong in the task, not on the final slide of a compliance course.
Workflow and human oversight establishes authority, handoffs, review, escalation, exception, workload effects, affected-party impact and recourse. Risk, ethics and social consequence addresses unfairness, exclusion, deception, unsafe use, legal duties, power and wider trade-offs. Change and operational stewardship covers behaviour monitoring, incident reporting, evidence preservation, skill updating and the point at which authority must be narrowed or suspended.
| Domain | Workplace question | Observable evidence |
|---|---|---|
| Recognition and boundaries | Where is AI acting, for what purpose and with which dependencies and limits? | Names the actual system, configuration, purpose, upstream sources and prohibited scope. |
| Task and tool judgment | Should AI be used, which mode fits, and what stays human? | Selects an approved route or a non-AI alternative and explains the boundary. |
| Interaction and specification | How should context, constraints, assumptions and iterations be expressed? | Creates a reproducible task specification without adding unlawful or unnecessary data. |
| Output and evidence judgment | What would support, contradict or qualify the output? | Checks claims, sources and uncertainty, documents reasoning and refuses unsupported work. |
| Data, privacy and security | Which information, rights, channels and manipulation risks apply? | Classifies data, uses the approved environment and detects leakage or injection traps. |
| Workflow and oversight | Who may act, review, escalate, override and provide recourse? | Uses the correct handoff and preserves meaningful human decision space. |
| Risk, ethics and consequence | Who could be excluded, deceived, harmed or denied rights? | Identifies population effects, duties and disproportionate consequences. |
| Change and stewardship | What must be monitored, reported, renewed or suspended? | Responds to incidents and material change with evidence, requalification or withdrawal. |
Role-Consequence Literacy MatrixSpecify capability at the point where authority meets impact
The Role-Consequence Literacy Matrix is this article's synthesis. For each role, name the work event and affected population; AI authority to observe, draft, recommend, decide, act, design, approve or challenge; consequence and reversibility; required depth across the eight domains; prohibited actions and mandatory escalation; representative scenarios; observable evidence and assessor; supervision and support; qualification expiry; and triggers for requalification.
Authority is more precise than a job title. A finance analyst might draft commentary in one workflow, recommend a forecast in another and approve no AI-generated material in a third. A manager may own the business outcome while lacking authority to change the system. An engineer may design orchestration but not approve use for a protected population. The matrix should be attached to the configured work event, not to a general identity such as “AI user” or “technical staff”.
Consequence includes severity, reach, reversibility, time pressure and the practical ability of an affected person to obtain review or correction. A frequent low-value drafting task can create cumulative disclosure risk. A rare approval can create a severe irreversible outcome. High consequence deepens evidence, oversight, data and stewardship requirements even when interaction technique is simple. Greater system autonomy can also move a role from checking output to governing action authority.
| Role event | Authority and consequence | Required evidence | Boundary |
|---|---|---|---|
| Internal email drafting | Draft only; low consequence; easily reversible | Approved channel, data classification, tone and factual check, non-AI option | No sensitive data, external commitment or automatic sending |
| Material financial forecast | Recommend; high decision consequence; correction may arrive late | Provenance, assumptions, reconciliation, uncertainty, challenge and escalation | No unsupported number enters executive or market-facing material |
| Customer recommendation | Recommend to adviser; population-dependent consequence | Cultural and legal edge cases, explanation, override, recourse and monitoring | No adviser treats confidence as authority or conceals system influence |
| Agent tool connection | Design action path; potentially high autonomy and exposure | Injection defence, least authority, logging, approvals, rollback and incident action | No unbounded credential, tool or downstream action |
| Deployment approval | Approve organisational authority; broad consequence | Population, purpose, evidence, human rights, failure boundary, owner and stop gate | No approval based on demo fluency or course participation |
| Independent challenge | Challenge without delivery ownership; consequence varies | Evaluation critique, incident analysis, control evidence and independent judgment | No ownership conflict or substitute product decision |
Five Qualification StatesGrant authority by person-role pairing, not by career rank
The Five Qualification States are this article's synthesis. Oriented means the person recognises AI, basic limits, approved-use rules and where to obtain help. Assisted practitioner means the person performs bounded tasks with templates, supervision and visible checking. Qualified operator means the person handles representative normal, ambiguous and failure scenarios within a declared authority. Consequential steward manages exceptions, affected-party duties, evidence, monitoring and change in a high-consequence domain. Independent challenger tests assumptions, evidence, controls and outcomes without owning delivery.
These states are not a career ladder. A senior executive can be oriented for hands-on tool use while qualified to approve a bounded deployment after demonstrating the relevant governance judgment. A data scientist can be a qualified operator for model evaluation but only an assisted practitioner in a regulated business workflow. A risk specialist may qualify as an independent challenger without being authorised to change the product. State belongs to a person-role-system relationship and carries an expiry and support condition.
Movement should follow protected practice. Orientation can enable low-consequence exploration in an approved environment. Assisted practitioners use templates and review while they encounter ambiguity and failure. Qualification requires observed action, not a perfect first attempt. Practice can include feedback, correction and replay before consequential authority is granted. A failed scenario should identify the unsafe pattern and the needed support rather than collapse into a permanent label.
Expiry follows material change, not an arbitrary annual certificate. Requalification is triggered when the system, model, tool, data class, autonomy, population, policy, role or consequence changes enough to invalidate prior evidence. Minor interface changes may require a briefing. A new action tool or protected population may require a full scenario. Incidents, repeated unsafe patterns or loss of support can narrow or suspend authority before the nominal date.
| State | Authority it can support | Proof and support | Typical next gate |
|---|---|---|---|
| Oriented | Recognise, ask, explore and use explicitly low-consequence approved features | Boundary recognition, approved-use rules, data hygiene and help route | Protected role practice |
| Assisted practitioner | Bounded draft or recommendation under visible supervision | Template use, checking, refusal and escalation in realistic practice | Observed qualification scenario |
| Qualified operator | Normal and exception work inside declared authority | Representative scenarios, evidence trail, correct handoffs and renewal trigger | Continue, deepen or narrow by consequence |
| Consequential steward | Manage high-consequence operation, exceptions, monitoring and change | Affected-party duties, evidence custody, incident action and requalification decisions | Independent challenge and operating review |
| Independent challenger | Challenge evidence, controls and outcomes without delivery ownership | Admissible critique, independence, escalation and clear limits on decision ownership | Assurance decision or remediation gate |
Scenario Qualification ProtocolObserve judgment where normal work becomes ambiguous
The Scenario Qualification Protocol is this article's synthesis. Use the actual role, approved tool, data class and workflow. Include a normal task, ambiguity, a confident error, missing evidence, a sensitive-data trap, a manipulation attempt, a population edge, an escalation and a credible non-AI alternative. Record the person's decision, reasoning, checks, evidence, refusal, escalation and outcome. The test asks whether the person preserves the declared boundary when the interface is persuasive and time is short.
A scenario should be representative without secretly recreating a live personnel decision. Synthetic or carefully protected cases can reproduce the decision conditions while avoiding unnecessary employee surveillance or exposure of customer data. Assessors need a rubric tied to observable work: correct system recognition, appropriate task choice, lawful context, evidence check, authority handling, affected-party protection and stewardship action. Personality, enthusiasm, typing speed and stylistic similarity to the assessor do not belong in the result.
Action and appropriate refusal carry equal weight. In a missing-evidence case, the qualified response may be to pause a forecast, request a source and escalate a deadline risk. In a manipulation case, it may be to isolate the input and report an incident. In a population edge, it may be to route the case to a specialist and explain the limit. A volume metric would punish these behaviours even though they preserve the organisation's control boundary.
| Scenario | What is observed | Unsafe pattern | Evidence retained |
|---|---|---|---|
| Normal and ambiguous work | System recognition, task choice, constraint handling and request for clarification | Using fluent output to conceal an unresolved assumption | Decision, reasoning, checks, result and assessor note |
| Confident error and missing evidence | Claim verification, uncertainty, reconciliation, refusal and escalation | Accepting confidence or inventing a citation | Sources inspected, contradiction found and disposition |
| Data trap and manipulation | Classification, approved channel, injection recognition and containment | Disclosing protected data or obeying untrusted instructions | Input handling, report, isolation and recovery action |
| Population edge and recourse | Fairness, context, explanation, override and affected-party route | Treating one population's experience as universal | Boundary, specialist input, communication and outcome |
| Escalation and non-AI alternative | Authority limit, workload awareness and selection of a safer route | Persisting with AI because use is rewarded | Handoff, support used, alternate workflow and next gate |
Literacy Evidence PassportCarry bounded authority without building a hidden worker score
The Literacy Evidence Passport is this article's synthesis. Create one record for each person-role pairing. It contains domain evidence, scenarios completed, observed strengths, unsafe patterns, current authority, prohibited actions, required support, assessor, qualification date, expiry, incidents, material changes and next practice. The passport answers an operational question: what may this person do in this configuration, under which supervision, until what event?
The record should remain legible to the worker. They should know what evidence was used, how to correct an error, how to demonstrate capability in another accessible form and how qualification affects authority. Learning data should have declared uses, access controls and retention. It must not become a hidden productivity, personality or promotion score. An observed unsafe pattern belongs to a role scenario with a support response, not to an enduring character judgment.
A material change updates the passport visibly. If an assistant gains action tools, a drafting qualification may become insufficient. If a high-consequence workflow removes automation and adds a specialist review, the required state may narrow. Incidents can trigger focused practice, interface repair or temporary restriction. Expiry is a control event linked to change and evidence, not a calendar ritual.
Literacy-to-Control BridgeRepeated failure may be a system diagnosis
The Literacy-to-Control Bridge is this article's synthesis. When people repeatedly fail a literacy scenario, do not assume that more instruction is the answer. Route the evidence through five possibilities: a knowledge or practice gap; an unsafe interface or weak default; authority wider than the role can exercise; workload or time pressure that makes checking impossible; or a workflow that withholds evidence, support or recourse.
Training is appropriate when practice changes performance under realistic conditions. An interface change is appropriate when the approved environment is hard to recognise, sensitive data is accepted without warning or sources are hidden behind fluent prose. Narrower authority is appropriate when a role cannot inspect the evidence needed to approve action. Specialist support is appropriate when a population or legal issue demands expertise that should not be distributed to every worker. Workflow redesign is appropriate when queues, targets or incentives punish checking and escalation.
Ten role-consequence casesQualify the work event, then change the control when necessary
Each case names the role, event, authority, consequence, domain requirement, scenario evidence, prohibited action, escalation, support, qualification state and control change. The purpose is not to create ten curricula. It is to show how one map produces different, minimum sufficient proof.
1. Approved assistant for internal email
Role and event: An employee drafts a low-consequence internal message with an approved assistant. Authority and consequence: Draft only; the employee remains the sender and can reverse the work before dispatch. Domains and scenario: Orientation emphasises system recognition, task choice, data classification and output checking. The scenario includes a confidential attachment and an invented meeting date. Evidence and boundary: The employee removes unnecessary data, corrects the date, preserves human ownership and uses a non-AI template when simpler. Automatic sending, external commitments and protected data are prohibited. Support and state: Approved examples, a channel marker and a help route support an oriented user. The next gate is assisted practice only if the role expands. Control change: Make the approved environment visually distinct and block sensitive attachments rather than teaching model engineering.
2. Material finance forecast
Role and event: A finance analyst prepares a forecast used in a material allocation decision. Authority and consequence: Recommend, with high consequence and delayed reversibility. Domains and scenario: The role needs task judgment, evidence, data governance, oversight and stewardship. A confident forecast cites a stale source, changes a hidden assumption and fails reconciliation. Evidence and boundary: The analyst traces provenance, exposes assumptions, compares uncertainty, reconciles the total and escalates the deadline. An unsupported number cannot enter executive material. Support and state: A supervisor, source register and finance validation template support supervised qualification before consequential use. Control change: Require machine-readable provenance and automatic reconciliation; training cannot compensate for a tool that hides sources.
3. Multilingual customer adviser
Role and event: An adviser receives recommendations for customers in several languages and cultures. Authority and consequence: Recommend and explain; a poor response can exclude a segment or misstate a legal right. Domains and scenario: Evidence, oversight, social consequence and escalation are deep requirements. One suggestion is fluent but culturally inappropriate and legally incomplete for a minority population. Evidence and boundary: The adviser recognises the edge, does not translate the error as settled advice, records the reason and routes to a specialist. Concealing AI influence or denying recourse is prohibited. Support and state: Localised guidance, specialist coverage and protected practice lead to supervised qualification. Control change: Add population-specific evaluation and an explicit specialist route rather than blaming language confidence.
4. Developer connecting tools to an agent
Role and event: A developer gives an agent access to records and a refund tool. Authority and consequence: Design, with potential autonomous action and broad exposure. Domains and scenario: System boundaries, security, oversight and stewardship are critical. A retrieved document contains an instruction to ignore policy and call the tool with elevated authority. Evidence and boundary: The developer recognises prompt injection, separates data from instruction, applies least privilege, logs the attempted action and demonstrates rollback. Unbounded credentials and silent action are prohibited. Support and state: Security review, tool sandbox and incident playbook support supervised qualification before consequential connection. Control change: Enforce scoped credentials and approval gates in the platform; awareness alone does not contain authority expansion.
5. Procurement leader evaluating vendor claims
Role and event: A procurement leader assesses claims about accuracy, integration and cost. Authority and consequence: Recommend shortlist and terms, with commercial and operational lock-in risk. Domains and scenario: Task judgment, evidence, data, risk and stewardship matter more than coding. A polished demo relies on supplier-selected cases and excludes reviewer cost. Evidence and boundary: The leader asks for buyer-held tests, population evidence, total outcome cost, dependency and exit proof. Vendor confidence and generic badges cannot substitute for witnessed evidence. Support and state: Technical, legal and domain specialists form an assessment team while the leader builds role-specific practice in a protected procurement exercise. Control change: Add non-compensable evidence gates and a portability test to the sourcing process.
6. Executive sponsor authorising deployment
Role and event: An executive sponsor approves deployment of a decision-support service. Authority and consequence: Approve organisational use across a consequential population. Domains and scenario: The sponsor needs boundaries, task judgment, oversight, consequence and stewardship, not prompt fluency. In the scenario, the business case is attractive but the sponsor cannot name the population, human authority, failure boundary, stop evidence or accountable owner. Evidence and boundary: Approval is withheld until these objects are explicit and independently challenged. A demo or course certificate cannot justify authority. Support and state: A risk challenger and capability owner pair with the sponsor. Control change: The approval paper must expose affected groups, recovery, evidence and stopping gates before the role can exercise approval.
7. Human-resources consequential workflow
Role and event: An HR team uses AI to assist a high-consequence employment workflow. Authority and consequence: Recommend and review; outcomes affect livelihood, disability accommodation and protected data. Domains and scenario: Data, evidence, oversight, ethics and affected-party recourse require depth. A candidate with an accessibility need is treated as anomalous and the explanation omits the relevant accommodation. Evidence and boundary: The team detects the population edge, pauses the route, seeks specialist input, documents the human decision and preserves contestability. Sole automated reliance and opaque employee scoring are prohibited. Support and state: Legal, accessibility and employee-representative input accompany supervised qualification. Control change: Narrow the system's role, require accessible alternatives and audit outcome differences.
8. Independent risk reviewer
Role and event: A reviewer critiques evaluation evidence, incidents and controls. Authority and consequence: Challenge without product ownership. Domains and scenario: Evidence, boundaries, risk, oversight and stewardship dominate. The product team presents an aggregate metric that hides one consequential population and treats absence of complaints as safety evidence. Evidence and boundary: The reviewer identifies the missing slice, tests admissibility, asks for recourse data and states the consequence of uncertainty. The reviewer must not quietly become the delivery decision owner. Support and state: Access to artefacts, protected independence and escalation qualify the person as an independent challenger. Control change: Preserve direct access to the accountable committee and separate remediation ownership from challenge.
9. Product steward after material change
Role and event: A product steward monitors drift, updates and user behaviour after a model and policy change. Authority and consequence: Manage operation and recommend restriction or requalification. Domains and scenario: Evidence, oversight and change stewardship are strongest. The new model improves average quality but changes refusals for one population and adds a tool. Evidence and boundary: The steward links behaviour to configuration, assesses affected roles, narrows action authority and triggers delta qualification. Treating the prior certificate as permanent is prohibited. Support and state: Monitoring, incident and workforce owners support consequential stewardship. Control change: Connect the change register to role passports so altered authority automatically opens the correct practice gate.
10. Repeated failures caused by the work system
Role and event: Workers repeatedly expose sensitive data and accept unsupported claims in a high-volume service. Authority and consequence: Draft and recommend under severe time pressure. Domains and scenario: Data and evidence scenarios show failure across trained groups. The approved tool resembles the public version, accepts restricted data, hides citations and leaves too little review time. Evidence and boundary: Workers can describe the rule but cannot perform it reliably within the interface and target. Blaming individuals or assigning another quiz is prohibited. Support and state: Authority is withheld until the control environment is ready. Control change: Separate the approved interface, block protected inputs, expose source state, lower workload, add specialist capacity and redesign the queue. Less training and a safer workflow are the right outcome.
AI Literacy ArchitectTurn role, consequence and observed evidence into one of five actions
The AI Literacy Architect is this article's synthesis. It returns exactly five outcomes: provide universal AI orientation; build role-specific practice in a protected environment; require supervised qualification before consequential use; pair the role with specialist or independent support; or withhold AI authority until the control environment is ready. It does not calculate a universal literacy score or infer capability from attendance.
Strong rival and failure boundaryMinimum sufficient qualification without a credential maze
The strongest rival deserves attention. Granular role-based literacy can become a bureaucratic credential system that slows access, expires too quickly and disadvantages workers with less formal education. Scenario assessment can encode cultural assumptions, reward rehearsal and drift away from real work. Fast-changing tools can make a curriculum stale before it reaches the workforce. Employers can also misuse learning records to rank people or transfer system failures into disciplinary files.
The practical answer is minimum sufficient qualification and portable foundations. Give everyone a concise common orientation. Add only the domain depth required by the actual authority and consequence. Use accessible work scenarios with more than one way to demonstrate judgment. Preserve evidence that transfers and reassess the delta after material change. Keep the passport visible, correctable and purpose-limited. Sample assessor consistency and investigate group differences before drawing conclusions about individuals.
Do not define literacy as prompt writing, give every role the same course or infer competence from seat time and quizzes. Product features without task judgment age quickly. Abstract warnings about fabricated output do little unless people practise checking and refusal. Privacy and security cannot be a closing slide. Authority, escalation and approved environments must exist before training grants live access. Technical specialists still need business and affected-party judgment; domain experts still need AI and data foundations.
Scope evidence carefully. The EU materials establish a current legal and implementation context for organisations within scope, not a universal certification model. UK, Canadian and OECD sources inform workplace practice and skill policy. UNESCO teacher and student frameworks are adapted here as examples of role-sensitive domains and progression, not transplanted as enterprise qualifications.89 NIST supports context, roles and lifecycle risk action. Research describes multidimensional literacy and task-oriented assessment, while the Eight-Domain Map, Role-Consequence Matrix, Five States, Scenario Protocol, Evidence Passport, Control Bridge and Architect remain this article's synthesis.1315
The final test is operational. Can the person recognise the system and boundary, choose AI or a non-AI route, provide lawful context, verify evidence, protect data, exercise only declared authority, account for affected people and respond to change? Can the organisation prove that the interface, workload, support and escalation make those behaviours possible? Literacy becomes meaningful when it grants a bounded permission that can be demonstrated, supported, narrowed and renewed.
Evidence spineSources checked 3 September 2026
- European Commission AI Office, AI Literacy: Questions and Answers.
- European Commission AI Office, Repository of AI Literacy Practices.
- European Union, Regulation (EU) 2024/1689, Artificial Intelligence Act.
- UK Government, AI Skills Tools Package.
- UK Government, Artificial Intelligence Playbook for the UK Government.
- Government of Canada, Guide on the Use of Generative Artificial Intelligence.
- OECD, Skill Needs and Policies in the Age of Artificial Intelligence.
- UNESCO, AI Competency Framework for Teachers.
- UNESCO, AI Competency Framework for Students.
- NIST, Artificial Intelligence Risk Management Framework.
- NIST, AI RMF Playbook.
- NIST, Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile.
- Long and Magerko, What Is AI Literacy? Competencies and Design Considerations.
- Zhang et al., Generative AI Literacy: A Comprehensive Framework for Literacy and Responsible Use.
- Bogart et al., AI Literacy Assessment Revisited: A Task-Oriented Approach Aligned with Real-world Occupations.
- Pinski and Benlian, Generative AI Literacy: Twelve Defining Competencies.
- Laupichler et al., Competency Model Approach to AI Literacy.
- Lyu et al., Generative Artificial Intelligence Literacy: Scale Development and Its Effect on Job Performance.
About this paperLiteracy is bounded permission backed by work evidence
The Eight-Domain AI Literacy Map, Role-Consequence Literacy Matrix, Five Qualification States, Scenario Qualification Protocol, Literacy Evidence Passport, Literacy-to-Control Bridge and AI Literacy Architect are this article's synthesis. They separate common orientation from the situated capability required to use, oversee, approve or challenge AI.