# XII. AI ## **12.1 AI-Era Competency Doctrine** ### **12.1.1 AI as Work Transformation, Not Only Tool Training.** 12.1.1.1 **SCF shall treat artificial intelligence as a structural transformation of work, learning, judgment, production, supervision, evidence, communication, risk management, and institutional capability, not merely as a tool-training subject.** AI-era competence shall therefore include not only the ability to use AI tools, but also the ability to understand how AI changes tasks, roles, workflows, accountability, human judgment, data stewardship, cybersecurity, public-safe communication, organizational design, labor-market transition, public authority learning, and lawful handoff. 12.1.1.2 SCF shall not reduce AI learning to prompt tips, software tutorials, automation shortcuts, productivity claims, vendor-specific tool adoption, or generic digital literacy. AI-era competence shall be designed as a layered capability system covering awareness, literacy, applied practice, supervised contribution, independent contribution, reviewer capability, maintainer capability, steward capability, mentor capability, systems leadership capability, national capability contribution, and handoff-context contribution. 12.1.1.3 AI-era competence shall include:\ (a) understanding AI as socio-technical infrastructure;\ (b) identifying where AI changes tasks, not merely job titles;\ (c) distinguishing augmentation, automation, delegation, decision support, simulation, retrieval, classification, summarization, generation, optimization, and agentic workflow;\ (d) recognizing data, privacy, cyber, bias, explainability, provenance, reliability, hallucination, dependency, and public-safe reporting risks;\ (e) applying human review and escalation;\ (f) recording AI-use labels, data-use labels, model cards, system cards, benchmark cards, and agent workflow cards where applicable;\ (g) preserving worker rights, public authority boundaries, procurement neutrality, finance-readiness boundaries, and non-execution discipline. 12.1.1.4 SCF shall ensure that AI-era learning and workforce transition do not overclaim that AI adoption alone creates competence, productivity, innovation, safety, readiness, financeability, procurement eligibility, public authority approval, or deployment authorization. ### **12.1.2 Human-AI Collaboration Competence.** 12.1.2.1 **Human-AI Collaboration Competence** shall mean the ability of a person, team, Competence Cell, National Working Group, public authority learning room, Foundry build team, Studio room, Academy cohort, Risk Academy cohort, or lawful downstream actor to work with AI systems while preserving human judgment, accountability, context, evidence quality, privacy, security, fairness, public-safe output, and correctionability. 12.1.2.2 Human-AI collaboration competence shall include the ability to:\ (a) select an appropriate AI-use mode;\ (b) define purpose and scope;\ (c) identify prohibited uses;\ (d) choose whether AI should be excluded, restricted, supervised, or permitted;\ (e) structure prompts, retrieval context, inputs, tools, and outputs safely;\ (f) validate outputs against evidence;\ (g) detect hallucination, bias, leakage, drift, overconfidence, and unsupported reasoning;\ (h) preserve human review;\ (i) escalate uncertain, high-risk, rights-affecting, public authority-sensitive, finance-sensitive, procurement-sensitive, or protected-knowledge-sensitive outputs;\ (j) record corrections and limitations. 12.1.2.3 Human-AI collaboration shall be treated as collaborative work under control, not as delegation of responsibility to a system. The human participant, reviewer, maintainer, steward, host, provider, employer, public authority, or lawful downstream recipient shall remain responsible within their applicable role and authority. ### **12.1.3 Automation Exposure Literacy.** 12.1.3.1 **Automation Exposure Literacy** shall mean the ability to understand how tasks, occupations, sectors, public-good workflows, National Portfolios, and handoff contexts may be exposed to automation, partial automation, augmentation, deskilling, displacement, role redesign, new supervision requirements, new data requirements, new cyber risks, and new accountability burdens. 12.1.3.2 SCF shall assess automation exposure at the task level before making claims at the occupation, sector, or workforce level. Exposure analysis shall distinguish:\ (a) tasks that may be automated;\ (b) tasks that may be augmented;\ (c) tasks requiring human judgment;\ (d) tasks requiring social trust or community relationship;\ (e) tasks requiring public authority decision-making;\ (f) tasks requiring professional licensure or regulated competence;\ (g) tasks involving protected knowledge or sensitive data;\ (h) tasks requiring field context, embodied practice, care, crisis response, or ethical judgment. 12.1.3.3 Automation exposure records shall not be used to declare that a worker, occupation, community, or sector is obsolete. SCF shall treat exposure as a transition signal, not as a displacement verdict. ### **12.1.4 AI Augmentation Literacy.** 12.1.4.1 **AI Augmentation Literacy** shall mean the ability to identify how AI can support human work without removing human accountability, degrading skill, increasing unsafe workload, intensifying surveillance, weakening labor protections, or producing unsupported authority claims. 12.1.4.2 AI augmentation may include assisted research, summarization, translation, coding, data cleaning, classification, drafting, visualization, scenario exploration, simulation support, risk-signal review, public-safe reporting support, learning support, accessibility support, and workflow triage. 12.1.4.3 AI augmentation shall require appropriate controls, including data minimization, purpose limitation, human review, output verification, prompt-injection awareness, source review, bias review, public-safe transformation, cybersecurity controls, and correction pathways. 12.1.4.4 AI augmentation shall not be treated as a basis for eliminating human judgment, reducing supervision, bypassing review, ignoring worker protection, or accelerating outputs beyond evidence and safeguard capacity. ### **12.1.5 AI Governance Literacy.** 12.1.5.1 **AI Governance Literacy** shall mean the ability to understand and apply the governance rules, records, labels, review gates, safeguard controls, risk classifications, data controls, human oversight requirements, incident procedures, and boundary rules that govern AI use within SCF and across Nexus. 12.1.5.2 AI Governance Literacy shall include:\ (a) AI-use labels;\ (b) data-use labels;\ (c) model card, system card, benchmark card, and agent workflow card literacy;\ (d) human-in-the-loop and human-on-the-loop concepts;\ (e) prohibited-use awareness;\ (f) high-stakes decision boundaries;\ (g) public authority boundary awareness;\ (h) finance, insurance, and procurement boundary awareness;\ (i) privacy and data protection;\ (j) cybersecurity and prompt-injection risk;\ (k) bias, discrimination, and harm review;\ (l) incident escalation and correction. 12.1.5.3 AI Governance Literacy shall be required for learners, mentors, reviewers, assessors, maintainers, Guild members, Competence Cell participants, WILP hosts, employer hosts, public authority learning participants, Foundry contributors, Studio participants, Risk Agency experts, and other actors using AI in SCF-related contexts. ### **12.1.6 AI Safety and Public-Safe Use.** 12.1.6.1 **AI Safety and Public-Safe Use** shall mean the safe, bounded, evidence-aware, rights-aware, privacy-protective, cyber-secure, non-misleading, and correctionable use of AI systems in learning, contribution, workforce transition, public-good production, public-safe reporting, research translation, Studio workflows, National Portfolios, Nexus Universe preparation, and handoff-context support. 12.1.6.2 Public-safe AI use shall avoid:\ (a) unsupported claims;\ (b) false certainty;\ (c) unverified risk warnings;\ (d) public authority overclaim;\ (e) financeability or insurability overclaim;\ (f) procurement implication;\ (g) consent overclaim;\ (h) protected knowledge exposure;\ (i) sensitive geospatial exposure;\ (j) data leakage;\ (k) discriminatory outputs;\ (l) automated high-stakes decisions. 12.1.6.3 AI safety and public-safe use shall require review, labels, source awareness, limitation statements, correction pathways, and escalation where outputs could affect rights, safety, public authority interpretation, employment, credentials, public trust, community standing, protected knowledge, or lawful handoff. ### **12.1.7 AI Work Redesign.** 12.1.7.1 **AI Work Redesign** shall mean the structured reconfiguration of tasks, workflows, roles, supervision, review, data use, tools, quality assurance, evidence production, human judgment, escalation, and correction in AI-enabled work environments. 12.1.7.2 SCF shall require AI work redesign to proceed by task decomposition, risk classification, human-AI allocation, review design, worker protection review, data governance review, cyber review, public-safe review, safeguard review, learning needs identification, competence mapping, WILP design where applicable, and correction planning. 12.1.7.3 AI work redesign shall not be used as a disguised labor-reduction claim, worker surveillance mechanism, unpaid productivity extraction tool, automated worker ranking system, or justification for unsafe delegation to AI. ### **12.1.8 AI Without Automated High-Stakes Decisions by Default.** 12.1.8.1 SCF shall prohibit automated high-stakes decisions by default. AI systems used in SCF shall not, by default, make or determine employment decisions, credential decisions, assessment decisions, public authority decisions, procurement decisions, finance decisions, insurance decisions, immigration decisions, disciplinary decisions, community-consent determinations, protected-knowledge permissions, public warnings, emergency commands, or deployment authorizations. 12.1.8.2 AI may support learning, drafting, classification, review, triage, summarization, retrieval, analysis, simulation, accessibility, translation, and quality improvement only where appropriate controls, human review, and boundary notices are applied. 12.1.8.3 Any proposed AI use that could influence high-stakes outcomes shall be escalated to appropriate governance, legal, ethical, data, AI, privacy, cyber, safeguard, public authority, credential, labor, or institutional review before use. *** ## **12.2 AI Skill Families** ### **12.2.1 AI Literacy.** 12.2.1.1 **AI Literacy** shall mean foundational understanding of what AI systems can and cannot do, including generative AI, predictive models, classification systems, retrieval systems, decision-support tools, agentic workflows, simulation models, digital twins, and AI-enabled automation. 12.2.1.2 AI Literacy shall include:\ (a) basic AI concepts;\ (b) data dependence;\ (c) probabilistic output awareness;\ (d) hallucination awareness;\ (e) bias and harm awareness;\ (f) privacy and confidentiality awareness;\ (g) cybersecurity awareness;\ (h) model limitation awareness;\ (i) human review awareness;\ (j) public-safe use awareness. 12.2.1.3 AI Literacy shall not be treated as AI professional competence, AI engineering competence, AI safety certification, or authorization to deploy AI systems. ### **12.2.2 Prompting and Interaction Skills.** 12.2.2.1 **Prompting and Interaction Skills** shall mean the ability to structure AI interactions to produce useful, bounded, reviewable, and safe outputs. 12.2.2.2 Prompting and interaction skills shall include purpose framing, context selection, instruction clarity, constraint setting, source citation requests where appropriate, output formatting, sensitivity awareness, role-boundary notices, hallucination checks, iterative refinement, and escalation. 12.2.2.3 Prompting skills shall include knowing when not to use AI, when to restrict inputs, when to avoid sensitive data, when to require secure-room use, when to require human expert review, and when to stop-the-line. 12.2.2.4 Prompting skill shall not be treated as domain expertise, professional judgment, evidence sufficiency, legal authority, public authority competence, or deployment readiness. ### **12.2.3 AI-Assisted Research Skills.** 12.2.3.1 **AI-Assisted Research Skills** shall mean the ability to use AI to support literature review, evidence mapping, research question formation, data exploration, method comparison, summary drafting, translation, coding support, and public-safe knowledge synthesis while preserving research integrity. 12.2.3.2 AI-assisted research shall require source review, citation discipline, method transparency, data provenance awareness, bias review, limitation statements, authorship transparency where appropriate, and correction pathways. 12.2.3.3 AI-assisted research outputs shall not be treated as validated research, expert review, peer review, public authority evidence, or publication-ready knowledge without human review and applicable governance. ### **12.2.4 AI-Assisted Coding and Automation Skills.** 12.2.4.1 **AI-Assisted Coding and Automation Skills** shall mean the ability to use AI systems to support software development, scripting, data processing, documentation, testing, debugging, refactoring, infrastructure-as-code drafting, notebook creation, API interaction, and workflow automation. 12.2.4.2 AI-assisted coding competence shall include secure coding awareness, dependency review, license review, secret scanning, test coverage, code review, vulnerability awareness, SBOM literacy, documentation review, reproducibility review, and no-warranty notice awareness. 12.2.4.3 AI-assisted code shall not be released, deployed, relied upon, published, or routed into public-good software, Studio workflows, DICE objects, Marketplace listings, Registry records, Grid inputs, TRL notes, or handoff contexts without appropriate human review and release controls. ### **12.2.5 AI Evaluation and Red Teaming Skills.** 12.2.5.1 **AI Evaluation and Red Teaming Skills** shall mean the ability to test, challenge, evaluate, document, and improve AI systems, outputs, prompts, workflows, tools, and model behavior under defined conditions. 12.2.5.2 Evaluation and red teaming may include accuracy checks, bias checks, safety testing, prompt-injection testing, jailbreak testing, data leakage testing, privacy testing, cyber misuse review, hallucination review, robustness testing, drift review, scenario testing, and public-safe output review. 12.2.5.3 Evaluation records shall include purpose, scope, test conditions, limitations, evaluator role, data basis, risk class, findings, unresolved issues, correction requirements, and archive. 12.2.5.4 Evaluation or red teaming shall not create AI safety certification, general validation, procurement readiness, financeability, insurability, public authority approval, or deployment authorization. ### **12.2.6 AI Governance and Risk Skills.** 12.2.6.1 **AI Governance and Risk Skills** shall include the ability to classify AI uses, identify risk levels, define controls, manage records, review outputs, handle incidents, apply public-safe constraints, preserve human oversight, and escalate high-risk cases. 12.2.6.2 These skills shall include knowledge of governance artifacts, including AI-use labels, data-use labels, model cards, system cards, benchmark cards, agent workflow cards, incident records, correction records, risk registers, assumption registers, dependency registers, and archive records. 12.2.6.3 AI governance and risk competence shall be required for participants who review, steward, maintain, publish, route, or display AI-enabled outputs within SCF. ### **12.2.7 Data Stewardship for AI.** 12.2.7.1 **Data Stewardship for AI** shall mean the ability to govern data used in or around AI systems, including training data, fine-tuning data, retrieval data, evaluation data, prompt inputs, output logs, user data, benchmark data, synthetic data, and metadata. 12.2.7.2 Data stewardship for AI shall include data provenance, rights review, consent and permission, data minimization, purpose limitation, sensitive data handling, youth data protection, health data protection, community data protection, Indigenous protocol-sensitive data controls where applicable, protected knowledge controls, cross-border controls, data sovereignty, and deletion, sealing, and archive. 12.2.7.3 Data stewardship competence shall include the ability to recognize when AI use is prohibited, restricted, secure-room-only, compute-to-data-only, public-safe-output-only, or subject to human review. ### **12.2.8 Model Card and System Card Literacy.** 12.2.8.1 **Model Card and System Card Literacy** shall mean the ability to read, create, review, interpret, and apply structured records describing AI models, AI systems, intended uses, prohibited uses, data basis, evaluation basis, limitations, risks, safeguards, human oversight requirements, support class, incident history, and correction pathways. 12.2.8.2 Model Card Literacy shall focus on model-level facts and limitations. System Card Literacy shall focus on system-level workflows, tool connections, human roles, data flows, runtime controls, access controls, and operational constraints. 12.2.8.3 Model cards and system cards shall be treated as records, not as certification, approval, warranty, public authority determination, or deployment authorization. ### **12.2.9 Agentic Workflow Literacy.** 12.2.9.1 **Agentic Workflow Literacy** shall mean the ability to understand and govern AI workflows that can plan, call tools, act across systems, retrieve data, write outputs, trigger actions, operate across steps, or interact with external environments. 12.2.9.2 Agentic workflow literacy shall include agent identity, tool permissions, human-in-the-loop controls, human-on-the-loop controls, no-command rules, no-write-back rules, output review, data leakage controls, prompt-injection controls, escalation triggers, kill-switch conditions, logging, and incident response. 12.2.9.3 Agentic workflows shall be prohibited from autonomous high-stakes action by default and shall not be used to execute public authority decisions, finance decisions, employment decisions, procurement decisions, protected-knowledge decisions, emergency commands, or deployment actions within SCF. ### **12.2.10 Human Oversight and Judgment Skills.** 12.2.10.1 **Human Oversight and Judgment Skills** shall mean the ability to interpret AI outputs, assess whether outputs are appropriate, identify errors and uncertainty, apply domain judgment, decide when to escalate, and prevent unsafe reliance. 12.2.10.2 Human oversight shall include review of facts, sources, assumptions, uncertainty, limitations, data sensitivity, public-safe status, legal boundaries, public authority implications, finance or procurement implications, and worker or community impacts. 12.2.10.3 Human judgment shall not be replaced by AI-generated confidence, fluency, ranking, scoring, summary, prediction, recommendation, or automation output. *** ## **12.3 Human Skills in AI-Augmented Work** ### **12.3.1 Analytical Thinking.** 12.3.1.1 SCF shall treat analytical thinking as a core AI-era competence. Participants shall be able to break down problems, identify variables, compare evidence, evaluate sources, test assumptions, distinguish signal from noise, and reason through uncertainty. 12.3.1.2 Analytical thinking shall be applied to AI outputs by checking claims, identifying unsupported conclusions, reviewing data basis, testing alternative explanations, and connecting outputs to recorded evidence. ### **12.3.2 Critical Thinking.** 12.3.2.1 Critical thinking shall include the ability to question AI outputs, detect false certainty, identify framing errors, recognize bias, evaluate incentives, challenge assumptions, and distinguish useful assistance from misleading automation. 12.3.2.2 Critical thinking shall be required before AI-assisted outputs are used in learning records, assessment, public-safe reporting, National Portfolios, Studio workflows, Registry records, Marketplace listings, Grid inputs, TRL notes, or handoff-context packages. ### **12.3.3 Creativity.** 12.3.3.1 Creativity shall remain a human capability in AI-augmented work. SCF shall support creative problem framing, scenario creation, design thinking, systems imagination, public-good innovation, community-sensitive design, and cross-domain synthesis. 12.3.3.2 AI may support creative exploration, but creative outputs shall be reviewed for originality, relevance, public-safe framing, cultural sensitivity, protected knowledge, and evidence basis. ### **12.3.4 Resilience and Adaptability.** 12.3.4.1 Resilience and adaptability shall include the ability to learn continuously, adjust to changing tools, respond to uncertainty, recover from errors, handle correction, move across roles, and maintain ethical judgment under pressure. 12.3.4.2 SCF shall treat adaptability as a supported capability, not as a demand that workers absorb unlimited disruption without institutional support, fair work, or learning time. ### **12.3.5 Ethics and Judgment.** 12.3.5.1 Ethics and judgment shall include the ability to identify harm, rights implications, discrimination risk, public trust concerns, community impacts, protected knowledge concerns, privacy risks, worker risks, public authority boundaries, and finance or procurement overclaims. 12.3.5.2 AI-era ethics shall be embedded in task design, data use, assessment, public display, output release, and lawful handoff context. ### **12.3.6 Communication and Sensemaking.** 12.3.6.1 Communication and sensemaking shall include the ability to explain AI-supported outputs, translate technical findings for different audiences, communicate uncertainty, avoid overclaim, prepare public-safe summaries, and distinguish evidence from speculation. 12.3.6.2 SCF shall train participants to communicate AI outputs with limitations, source context, review status, and boundary notices. ### **12.3.7 Collaboration.** 12.3.7.1 Collaboration shall include the ability to work across human teams, AI-assisted workflows, public-good institutions, national actors, communities, public authorities, technical experts, reviewers, mentors, and lawful downstream actors. 12.3.7.2 AI-supported collaboration shall preserve role clarity, attribution, review responsibility, data controls, and correction pathways. ### **12.3.8 Leadership.** 12.3.8.1 AI-era leadership shall include the ability to guide responsible AI adoption, protect workers, preserve evidence quality, maintain public trust, prevent automation overclaim, support learning, and govern human-AI work redesign. 12.3.8.2 Leadership shall include accountability for what AI is allowed to do, what humans must review, what data may be used, what outputs may be published, what incidents must be escalated, and what boundaries must be preserved. ### **12.3.9 Active Listening and Contextual Understanding.** 12.3.9.1 Active listening and contextual understanding shall be required for AI-era work involving communities, public authorities, workers, learners, Indigenous participants where applicable, vulnerable groups, disaster-affected populations, and sector experts. 12.3.9.2 AI-generated summaries or interpretations shall not replace direct listening, contextual understanding, local knowledge, community protocols, professional judgment, or public authority process. ### **12.3.10 Socio-Technical Systems Thinking.** 12.3.10.1 Socio-technical systems thinking shall mean the ability to understand how people, institutions, technologies, data, law, infrastructure, incentives, labor markets, culture, risks, safeguards, and public-good records interact. 12.3.10.2 SCF shall treat socio-technical systems thinking as essential for AI-era competence because AI systems can alter not only tasks, but also authority, trust, accountability, labor relations, public meaning, and institutional risk. *** ## **12.4 Work Redesign** ### **12.4.1 Task Decomposition.** 12.4.1.1 **Task Decomposition** shall be the required starting point for AI work redesign. SCF shall decompose roles into tasks, sub-tasks, decisions, judgments, inputs, outputs, tools, dependencies, risks, review points, data requirements, and human accountability points before determining whether AI may be used. 12.4.1.2 Task decomposition shall identify:\ (a) routine tasks;\ (b) judgment-intensive tasks;\ (c) safety-critical tasks;\ (d) rights-affecting tasks;\ (e) public authority-sensitive tasks;\ (f) finance, insurance, or procurement-sensitive tasks;\ (g) protected knowledge-sensitive tasks;\ (h) community-sensitive tasks;\ (i) data-sensitive tasks;\ (j) tasks suitable for learning, simulation, or controlled practice only. 12.4.1.3 Task decomposition shall not be used to fragment work into extractive micro-labor, deskill workers, hide accountability, or justify unsafe automation. ### **12.4.2 Human-in-the-Loop Workflows.** 12.4.2.1 **Human-in-the-Loop Workflows** shall require human review, approval, or intervention before an AI-supported output can move to the next stage, be recorded, be displayed, be routed, be published, be used in assessment, or be included in a handoff-context package. 12.4.2.2 Human-in-the-loop controls shall be required for AI outputs involving sensitive data, public-safe reporting, credential evidence, learner assessment, worker profiles, risk intelligence, public authority learning, finance-readiness, procurement boundaries, protected knowledge, safeguard records, or lawful handoff. 12.4.2.3 Human-in-the-loop review shall be meaningful, not symbolic. Reviewers must have appropriate competence, access to evidence, authority to reject or correct, and a recorded correction pathway. ### **12.4.3 Human-on-the-Loop Oversight.** 12.4.3.1 **Human-on-the-Loop Oversight** shall mean ongoing supervision of AI-assisted workflows where the system may generate, triage, classify, summarize, or recommend within predefined boundaries, but a human steward monitors performance, detects drift, reviews exceptions, and can intervene. 12.4.3.2 Human-on-the-loop oversight may be appropriate for lower-risk workflows, provided that the system cannot make high-stakes determinations, write back to authoritative systems, trigger operational commands, or bypass review gates. 12.4.3.3 Human-on-the-loop oversight shall include logs, review sampling, escalation triggers, incident reporting, correction, and periodic reauthorization. ### **12.4.4 AI Co-Pilot Workflows.** 12.4.4.1 **AI Co-Pilot Workflows** shall mean AI-assisted workflows where AI supports a human participant in drafting, analysis, coding, research, learning, translation, visualization, simulation, or review while the human remains responsible for judgment, verification, and submission. 12.4.4.2 AI co-pilot use shall be allowed only within the relevant data, AI-use, privacy, cyber, public-safe, and safeguard controls. 12.4.4.3 Co-pilot output shall not be treated as reviewed, accurate, complete, evidence-sufficient, or publishable unless independently checked under applicable workflow rules. ### **12.4.5 Agentic Task Controls.** 12.4.5.1 **Agentic Task Controls** shall apply where AI systems can perform multi-step tasks, use tools, retrieve information, initiate actions, generate files, manipulate data, query systems, or interact with external services. 12.4.5.2 Agentic workflows shall include predefined tool permissions, no-command rules, no-write-back rules, data access controls, output review, logging, escalation triggers, kill-switch conditions, and incident response. 12.4.5.3 Agentic workflows shall not autonomously publish, certify, approve, reject, rank, hire, fire, procure, finance, insure, grant consent, issue warnings, command operations, deploy systems, or make public authority actions within SCF. ### **12.4.6 Quality Assurance and Review.** 12.4.6.1 AI work redesign shall include quality assurance and review controls proportionate to risk. Controls may include peer review, mentor review, expert review, technical review, data review, AI-use review, cyber review, privacy review, public-safe review, safeguard review, and legal boundary review. 12.4.6.2 AI-supported work products shall carry review status, evidence basis, AI-use label, data-use label, limitations, correction pathway, and archive rule where applicable. 12.4.6.3 Quality assurance shall not be replaced by AI-generated confidence, automated scoring, fluency, benchmark claims, or unverified synthetic evidence. ### **12.4.7 Error Escalation.** 12.4.7.1 AI-enabled workflows shall include error escalation where outputs are incorrect, uncertain, biased, harmful, misleading, unsupported, privacy-compromising, cyber-sensitive, public authority-sensitive, finance-sensitive, procurement-sensitive, protected-knowledge-sensitive, or public-safe-risk-bearing. 12.4.7.2 Error escalation may trigger correction, review, reclassification, withdrawal, suspension, stop-the-line, public-safe notice, record update, reviewer reassignment, model withdrawal, workflow restriction, or archive. 12.4.7.3 Error escalation shall protect learners, workers, contributors, communities, and public trust from silent propagation of AI errors. ### **12.4.8 Bias and Harm Review.** 12.4.8.1 AI work redesign shall include bias and harm review where AI outputs may affect learning, assessment, credentialing, worker profiles, public display, labor-market interpretation, community representation, risk intelligence, public authority learning, or handoff context. 12.4.8.2 Bias and harm review shall consider protected characteristics, disability, gender, youth, rurality, language, migration status, community identity, Indigenous protocols where applicable, socioeconomic status, disaster exposure, conflict exposure, and historical exclusion. 12.4.8.3 Bias and harm review shall include correction, appeal, public repair where necessary, and prevention of repeated harm. ### **12.4.9 Productivity Without Worker Exploitation.** 12.4.9.1 SCF shall treat AI-enabled productivity as legitimate only where it does not produce worker exploitation, unreasonable workload intensification, surveillance expansion, unpaid labor extraction, deskilling, hidden displacement, unsafe automation, harassment, discrimination, or loss of worker agency. 12.4.9.2 AI productivity claims shall be evidence-based, bounded, context-specific, and reviewed for job-quality effects. 12.4.9.3 SCF shall not use AI productivity metrics as worker ranking, wage suppression, automated discipline, or employment decision tools by default. ### **12.4.10 No Automated Worker Displacement Claims by SCF.** 12.4.10.1 SCF shall not automatically declare workers, roles, occupations, or communities displaced, obsolete, redundant, or substitutable based on AI analysis, labor-market forecasts, automation exposure, task mapping, or vendor claims. 12.4.10.2 Displacement risk shall be recorded as a transition signal requiring human review, labor-market context, worker voice, employer context, sector context, public authority context, and rights-aware interpretation. 12.4.10.3 Any workforce restructuring, redundancy, layoff, redeployment, or employment decision shall occur outside SCF through competent lawful actors and applicable labor, employment, union, public authority, and institutional processes. *** ## **12.5 AI Labor-Market Monitoring** ### **12.5.1 AI-Exposed Occupations.** 12.5.1.1 SCF may monitor AI-exposed occupations by identifying tasks, workflows, decision points, data dependencies, routine activities, cognitive activities, clerical activities, analytical activities, creative activities, coding activities, service activities, professional activities, and public-sector activities that may be affected by AI. 12.5.1.2 AI exposure shall be recorded with uncertainty, source limitations, sector context, jurisdictional context, job-quality context, and human review. 12.5.1.3 AI exposure shall not be interpreted as inevitable displacement or readiness for automation. ### **12.5.2 AI-Augmented Occupations.** 12.5.2.1 SCF may identify AI-augmented occupations where AI can support existing human work through research, summarization, drafting, coding, translation, data work, visualization, scenario analysis, risk interpretation, learning, accessibility, and workflow support. 12.5.2.2 AI augmentation records shall identify required skills, human review needs, data governance requirements, cyber controls, public-safe requirements, and worker protection considerations. 12.5.2.3 AI augmentation shall not be treated as proof that a worker can safely perform a role without training, supervision, domain competence, or review. ### **12.5.3 AI-Disrupted Tasks.** 12.5.3.1 SCF shall monitor AI-disrupted tasks where AI changes task frequency, task value, human effort, required competence, supervision model, quality expectations, data requirements, or risk profile. 12.5.3.2 AI-disrupted task records shall be connected to bridge competencies, short-cycle reskilling, upskilling, WILPs, Micro-Credentials, and transition supports. 12.5.3.3 AI-disrupted task records shall not be used as automated evidence of individual worker deficiency. ### **12.5.4 Emerging AI Roles.** 12.5.4.1 SCF may identify emerging AI roles, including AI literacy facilitator, AI workflow reviewer, AI governance analyst, model card steward, system card steward, agent workflow reviewer, AI safety reviewer, prompt-injection tester, AI red-team contributor, AI data steward, human oversight lead, AI public-safe reporting reviewer, and AI-enabled workflow maintainer. 12.5.4.2 Emerging AI roles shall be mapped to competencies, evidence types, learning pathways, WILPs, Micro-Credentials, mentor requirements, review requirements, and boundary rules. 12.5.4.3 Emerging role recognition shall not create professional licensure, employment status, procurement qualification, public authority recognition, or deployment authority by default. ### **12.5.5 Declining AI-Substitutable Tasks.** 12.5.5.1 SCF may monitor tasks that appear increasingly substitutable by AI or automation, provided such monitoring is conducted with uncertainty, human review, labor-market context, worker protection, and rights-aware interpretation. 12.5.5.2 Declining task signals shall be used to design transition supports, not to stigmatize workers or justify unmanaged displacement. 12.5.5.3 Declining task records shall include alternative pathways, bridge competencies, reskilling options, and job-quality considerations where possible. ### **12.5.6 Entry-Level Role Transformation.** 12.5.6.1 SCF shall monitor how AI changes entry-level roles, including reduced routine task availability, increased expectations for AI fluency, shifted apprenticeship models, changed supervision, higher portfolio expectations, and new evidence requirements. 12.5.6.2 SCF shall design entry-level pathways that protect early-career workers from losing practice opportunities because routine tasks have been automated without replacement learning structures. 12.5.6.3 Entry-level transformation shall be addressed through WILPs, simulations, Studio exercises, Foundry starter tasks, supervised contribution, mentoring, Micro-Credentials, and portfolio evidence. ### **12.5.7 AI Skills Premium Monitoring.** 12.5.7.1 SCF may monitor whether AI skills appear associated with wage premiums, demand premiums, hiring signals, promotion signals, or sector mobility signals. 12.5.7.2 AI skills premium monitoring shall include source limitations, labor-market variation, sector variation, geographic variation, occupational variation, equity effects, credential inflation risks, and time sensitivity. 12.5.7.3 AI skills premium monitoring shall not be represented as wage promise, salary advice, employment guarantee, or guaranteed return on learning investment. ### **12.5.8 AI Equity and Access Monitoring.** 12.5.8.1 SCF shall monitor AI equity and access, including disparities in tool access, connectivity, language, disability access, educational access, employer access, safe practice environments, mentorship, data access, and opportunity to build AI-era portfolios. 12.5.8.2 AI equity monitoring shall identify barriers affecting women, youth, people with disabilities, rural and remote communities, migrants, refugees, informal workers, gig workers, underrepresented groups, climate-affected workers, disaster-affected workers, and Indigenous participants where applicable. 12.5.8.3 AI equity monitoring shall be used for inclusion improvement, not ranking, profiling, or social scoring. ### **12.5.9 AI Credential Inflation Monitoring.** 12.5.9.1 SCF shall monitor AI credential inflation, including situations where employers, platforms, providers, education institutions, or public-good programs demand AI credentials beyond what a role actually requires. 12.5.9.2 Credential inflation monitoring shall distinguish genuine competence requirements from exclusionary credential requirements, marketing claims, vendor capture, and artificial barriers to entry. 12.5.9.3 SCF shall avoid contributing to credential inflation by ensuring that Micro-Credentials are bounded, evidence-based, role-relevant, correctable, and not overclaimed. ### **12.5.10 AI Skill Obsolescence Monitoring.** 12.5.10.1 SCF shall monitor AI skill obsolescence because AI tools, models, interfaces, risks, and governance expectations may change rapidly. 12.5.10.2 AI skill records, Micro-Credentials, learning pathways, and portfolio displays may include expiry, renewal, refresh, revision, correction, or archive rules. 12.5.10.3 Obsolescence monitoring shall not invalidate a worker’s broader competence by default; it shall identify refresh needs for specific tools, workflows, models, or risk controls. *** ## **12.6 AI Work Safeguards** ### **12.6.1 Data Privacy.** 12.6.1.1 AI work shall comply with SCF data privacy controls, including data minimization, purpose limitation, consent and permission, sensitive data handling, learner data protection, worker data protection, youth data protection, health data protection, community data protection, Indigenous protocol-sensitive data controls where applicable, cross-border controls, deletion, sealing, archive, and correction. 12.6.1.2 Participants shall not enter confidential, restricted, personal, protected, public authority-sensitive, health-sensitive, youth-sensitive, cyber-sensitive, infrastructure-sensitive, community-sensitive, Indigenous protocol-sensitive, or protected knowledge into AI systems unless permitted by the applicable AI-use label, data-use label, access class, room control, and review process. ### **12.6.2 Worker Surveillance Limits.** 12.6.2.1 SCF shall prohibit AI-enabled surveillance of learners, workers, contributors, mentors, reviewers, assessors, WILP participants, or Competence Cell participants unless specifically authorized for a legitimate, proportionate, disclosed, limited, and rights-protective purpose. 12.6.2.2 AI monitoring shall not be used for hidden productivity scoring, automated discipline, wage suppression, unsafe workload intensification, social scoring, discrimination, or retaliation. 12.6.2.3 Any monitoring required for security, safety, learning integrity, credential integrity, or platform integrity shall be minimized, logged, reviewed, and subject to correction and grievance. ### **12.6.3 Algorithmic Management Literacy.** 12.6.3.1 SCF shall include algorithmic management literacy so participants understand automated scheduling, task allocation, platform ratings, productivity tracking, automated scoring, automated discipline, AI hiring tools, AI performance management, and related worker risks. 12.6.3.2 Algorithmic management literacy shall help participants recognize rights issues, fairness issues, data protection issues, explainability issues, bias issues, and grievance needs. 12.6.3.3 SCF shall not deploy algorithmic management systems to rank, discipline, exclude, or manage participants by default. ### **12.6.4 Bias and Discrimination Controls.** 12.6.4.1 AI work safeguards shall include bias and discrimination controls across learning, assessment, credentialing, worker profiles, labor-market monitoring, portfolio display, mentor review, employer-readable profiles, and public-safe reporting. 12.6.4.2 Bias controls shall address representational bias, measurement bias, historical bias, language bias, disability bias, gender bias, racial or ethnic bias, geographic bias, socioeconomic bias, migration-status bias, and bias affecting disaster-affected, climate-affected, rural, Indigenous, or underrepresented populations. 12.6.4.3 Where bias or discrimination is identified, SCF shall support correction, appeal, public repair where necessary, record revision, model or workflow restriction, retraining of reviewers where applicable, and archive. ### **12.6.5 Explainability Literacy.** 12.6.5.1 SCF shall include explainability literacy so participants can understand when an AI output can be explained, when it cannot, what type of explanation is available, what explanation is sufficient for the workflow, and when lack of explanation requires restriction or escalation. 12.6.5.2 Explainability literacy shall include distinction between technical explanations, user-facing explanations, policy explanations, legal explanations, public-safe explanations, and community-facing explanations. 12.6.5.3 Lack of explainability shall be a risk factor for high-stakes, public authority-sensitive, rights-affecting, credentialing, employment, finance, insurance, procurement, and community-sensitive uses. ### **12.6.6 Cybersecurity Controls.** 12.6.6.1 AI work safeguards shall include cybersecurity controls, including prompt-injection awareness, data leakage prevention, tool permission controls, identity and access management, least privilege, secure repositories, secret scanning, dependency scanning, vulnerability awareness, output sanitation, logging, and incident response. 12.6.6.2 AI workflows connected to tools, repositories, APIs, secure rooms, data rooms, Studio environments, Registry systems, Marketplace systems, or handoff-context packages shall require additional cyber review. 12.6.6.3 AI systems shall not be allowed to execute commands, modify records, publish outputs, access restricted systems, or trigger external actions without explicit authorization and human review. ### **12.6.7 Human Review Requirements.** 12.6.7.1 Human review shall be required for AI outputs used in assessment, Micro-Credentials, Skills Wallets, ILA records, employer-readable profiles, Registry records, Marketplace listings, Reports, public-safe summaries, Studio workflows, Grid inputs, TRL notes, National Portfolios, public authority learning records, finance-readiness notes, procurement boundary notes, and handoff-context packages. 12.6.7.2 Human review shall include competence-appropriate review, evidence review, source review, limitation review, sensitivity review, public-safe review, and correction review. 12.6.7.3 Human review shall be recorded where the output is used as an evidence-bearing or public-facing object. ### **12.6.8 Public Authority Boundary Controls.** 12.6.8.1 AI shall not be used within SCF to make public authority decisions, determine public benefits, issue public warnings, approve public finance, grant permits, approve procurement, make regulatory determinations, command emergency response, or represent official policy. 12.6.8.2 AI may support public authority learning only where the learning status, non-decision status, no-warning status, no-approval status, and no-substitution status are clear. 12.6.8.3 Public authority-sensitive AI outputs shall require public authority boundary review and public-safe transformation before external use. ### **12.6.9 No AI-Based Employment Decision by SCF.** 12.6.9.1 SCF shall not use AI to make employment decisions, hiring decisions, firing decisions, promotion decisions, wage decisions, worker ranking decisions, eligibility decisions, redeployment decisions, disciplinary decisions, or exclusion decisions. 12.6.9.2 AI may assist participants in understanding skills, learning options, portfolio gaps, transition pathways, and labor-market signals, provided that outputs are advisory, bounded, reviewed where necessary, and not used as employment determinations. 12.6.9.3 Employers, hosts, and lawful downstream actors remain responsible for lawful, fair, non-discriminatory, human-reviewed employment and engagement decisions outside SCF. ### **12.6.10 AI Incident Correction.** 12.6.10.1 AI incidents shall include hallucination incidents, bias incidents, discrimination incidents, data leakage incidents, privacy incidents, cyber incidents, prompt-injection incidents, agentic workflow incidents, unsafe output incidents, public-safe reporting incidents, credentialing incidents, worker-profile incidents, public authority boundary incidents, finance boundary incidents, procurement boundary incidents, consent overclaim incidents, and handoff overclaim incidents. 12.6.10.2 AI incident correction may include containment, claims freeze, data freeze, technical freeze, model freeze, workflow suspension, output withdrawal, public-safe notice, record correction, reassessment, credential correction, Marketplace delisting, Registry update, handoff recall, archive, and non-continuation. 12.6.10.3 AI incident correction shall propagate to affected records, outputs, learners, workers, contributors, public-safe materials, Registry entries, Marketplace listings, Reports, Studio workflows, Grid inputs, TRL notes, and handoff-context packages where applicable. *** ## **12.7 AI Boundary Rules** ### **12.7.1 AI Skill Is Not AI Safety Certification.** 12.7.1.1 No AI skill record, AI Micro-Credential, AI learning object, AI WILP, AI portfolio artifact, AI iCRS record, AI Foundry contribution, AI Studio exercise, or AI assessment record shall constitute AI safety certification by default. 12.7.1.2 AI safety certification, where applicable, must be issued separately by a competent authority or qualified body under applicable rules and scope. ### **12.7.2 AI Tool Use Is Not Professional Competence Alone.** 12.7.2.1 Use of an AI tool shall not, by itself, demonstrate professional competence, domain competence, public authority competence, technical competence, legal competence, clinical competence, engineering competence, emergency management competence, or handoff readiness. 12.7.2.2 Professional competence requires evidence, context, judgment, review, applicable standards, lawful authority where required, and role-specific assessment. ### **12.7.3 AI Assessment Is Not Hiring Decision.** 12.7.3.1 AI-assisted assessment, skills analysis, portfolio review, credential review, or pathway recommendation within SCF shall not constitute a hiring decision, employment eligibility determination, promotion decision, redeployment decision, wage decision, or worker ranking. 12.7.3.2 Any employment-related use of SCF records by employers or hosts shall require independent lawful process, human review, non-discrimination controls, privacy controls, and applicable worker protections. ### **12.7.4 AI Forecast Is Not Labor-Market Certainty.** 12.7.4.1 AI-generated labor-market forecasts, automation exposure analyses, AI skills demand signals, wage context, emerging role analyses, and transition pathway recommendations shall not be treated as certainty. 12.7.4.2 Such outputs shall carry uncertainty, source limitations, geographic context, sector context, time sensitivity, and human review status. ### **12.7.5 AI Credential Is Not Deployment Approval.** 12.7.5.1 AI-related Micro-Credentials, badges, learning records, Skills Wallet entries, WILP records, or portfolio artifacts shall not authorize deployment of AI systems, use of AI in high-stakes contexts, production release, public authority use, procurement use, finance use, insurance use, or operational command. 12.7.5.2 Deployment approval, where applicable, must be made separately by competent lawful actors under applicable governance, legal, technical, safety, cyber, data, and public authority processes. ### **12.7.6 SCF Does Not Automate Worker Ranking by Default.** 12.7.6.1 SCF shall not automate worker ranking, learner ranking, contributor ranking, employability scoring, social scoring, wage scoring, risk scoring, productivity scoring, or public display ranking by default. 12.7.6.2 SCF metrics, Skills Wallet records, ILA records, Micro-Credentials, iCRS records, WILP records, and portfolio displays shall support learning, evidence, transition, recognition, and correction, not automated exclusion or hierarchy. *** ## **12.8 Final Part XII Operating Statement** 12.8.1 SCF shall treat AI-era work as a transformation of tasks, workflows, judgment, learning, production, evidence, labor markets, public authority learning, national capability, and lawful handoff. It shall not reduce AI competence to tool training, prompt use, vendor familiarity, or productivity claims. 12.8.2 SCF shall structure AI-era competence through AI literacy, prompting and interaction skills, AI-assisted research, AI-assisted coding and automation, AI evaluation and red teaming, AI governance and risk skills, data stewardship for AI, model card and system card literacy, agentic workflow literacy, and human oversight and judgment skills. 12.8.3 SCF shall preserve the central role of human skills in AI-augmented work, including analytical thinking, critical thinking, creativity, resilience, adaptability, ethics, judgment, communication, sensemaking, collaboration, leadership, active listening, contextual understanding, and socio-technical systems thinking. 12.8.4 SCF shall govern AI work redesign through task decomposition, human-in-the-loop workflows, human-on-the-loop oversight, AI co-pilot workflows, agentic task controls, quality assurance, error escalation, bias and harm review, productivity without worker exploitation, and prohibition of automated worker displacement claims by SCF. 12.8.5 SCF shall monitor AI labor-market change through AI-exposed occupations, AI-augmented occupations, AI-disrupted tasks, emerging AI roles, declining AI-substitutable tasks, entry-level role transformation, AI skills premium monitoring, AI equity and access monitoring, AI credential inflation monitoring, and AI skill obsolescence monitoring, always with uncertainty, human review, worker protection, and no employment or wage guarantees. 12.8.6 SCF shall apply AI work safeguards for data privacy, worker surveillance limits, algorithmic management literacy, bias and discrimination controls, explainability literacy, cybersecurity controls, human review requirements, public authority boundary controls, no AI-based employment decisions by SCF, and AI incident correction. 12.8.7 The final rule of Part XII is that AI shall support learning, evidence, contribution, capability, public-good production, workforce resilience, and lawful handoff context only within bounded, human-reviewed, privacy-protective, cyber-secure, public-safe, correctionable, and non-executing controls. AI skill is not AI safety certification; AI tool use is not professional competence alone; AI assessment is not hiring decision; AI forecast is not labor-market certainty; AI credential is not deployment approval; and SCF shall not automate worker ranking by default. --- # Agent Instructions: Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter: ``` GET https://docs.therisk.global/organization/operations/frameworks/sustainable-competency-framework-scf/xii.-ai.md?ask= ``` The question should be specific, self-contained, and written in natural language. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.