What the U.S. Department of Labor AI Framework Asks Workers to Learn

What is DOL asking workers to learn—and is it a requirement?

The U.S. Department of Labor’s February 13, 2026 framework asks workforce and education programs to help people understand AI, explore uses, direct systems, evaluate outputs, and use AI responsibly. It is a flexible resource for designing learning, not a federal checklist each worker must complete. In its accompanying notice, DOL describes the framework as voluntary guidance and encourages workforce and education systems to use it in program design. [The U.S. Department of Labor’s Artificial Intelligence Literacy Framework](https://rosap.ntl.bts.gov/view/dot/93248/dot_93248_DS1.pdf) therefore sets a learning direction; it does not establish that a particular employer has adopted AI, predict which jobs will change, or tell an individual whether their job is secure. Its stated audience includes workers, employers, training providers, educators, and public workforce agencies. That breadth matters: the document guides program design, while role-specific expectations remain contextual. For a worker, the practical starting point is a real goal or recurring task, plus the employer’s rules for approved tools and data. Choose that context before buying a course; the framework itself does not prescribe one universal syllabus, tool, or career move.

Sources: The U.S. Department of Labor's Artificial Intelligence Literacy Framework (ROSA P record); Attachment I (Accessible PDF): The U.S. Department of Labor's Artificial Intelligence Literacy Framework; Training and Employment Guidance Letter No. 10-25

What should a worker be able to do in each of the five areas?

A worker who has learned the five areas should be able to explain what an AI system may do and where it can fail, recognize a task where it might help, give it relevant context, check its output against a sound standard, and follow the rules that govern its use. It is a baseline, not a demand to build a model or memorize prompts. The Department of Labor’s *Artificial Intelligence Literacy Framework* describes broad foundations for using and evaluating AI responsibly. It distinguishes foundational literacy from advanced work such as managing or building AI systems, with depth shaped by role and context. Understanding AI principles means having a useful mental model rather than technical mastery. Generative systems produce outputs from learned patterns; the same request can yield different responses, and confident wording can be wrong. For everyday work, the consequence is simple: treat a response as a proposal to inspect, not as a fact whose fluency proves accuracy. Learn what the system has and has not established for the task, not the mathematics of model training. Exploring AI uses means connecting capabilities to real work. Ask which part of the worker’s task might benefit, given its purpose, inputs, and consequences. Applications vary across industries and roles. A first draft may be worth exploring; consequential decisions need stronger controls or no AI use. Identify the job step and its quality standard, then ask whether AI would be useful and permitted. Tool familiarity alone does not establish workflow fit. Directing AI effectively means giving it relevant context so the result can be assessed. DOL includes framing the task, specifying audience and goals, supplying appropriate supporting material, and iterating when an initial output misses the need. Better direction does not guarantee correctness, so evaluation remains necessary. Evaluating outputs means checking accuracy, completeness, logic, clarity, and fit against a standard outside the output itself. It calls for cross-checking claims against trusted sources, noticing omissions or flawed assumptions, and applying expertise. A summary can be checked against its source; figures and calculations against underlying data. A style preference cannot verify a factual claim, and a plausible answer is not its own evidence. Verification also has a cost. If checking takes longer than doing the task directly, or the worker cannot verify the result, the proposed use may not help. Responsible use keeps the other areas within workplace boundaries. DOL names protecting sensitive information, following organizational rules, avoiding misuse or harm, considering context-specific risk, and maintaining accountability. A worker needs to know approved tools, data limits, review ownership, and how to report concerns. An internal draft and a decision affecting money, access, safety, or personal data need different scrutiny. AI is not the accountable decision-maker; baseline literacy does not authorize a worker to override policy or approve high-stakes workflows. If data boundaries or review ownership are unclear, asking before use is responsible practice. Together, the areas form a useful learning cycle: understand limitations, find a task-specific use, provide permitted context, evaluate the result, and decide whether and how to use it. This is an interpretation, not a required sequence. A learner can explain the task, show the instruction and check, and name the remaining human responsibility. This shows practical learning, not readiness for a different occupation. The durable learning lies in understanding the task, bringing domain knowledge, framing a problem, checking evidence, and communicating limits. Interfaces and prompt conventions can change; these habits still help a worker assess a new tool. For current-role use, a checked work sample can demonstrate learning. A person aiming to build software or pursue machine-learning engineering or research needs deeper technical preparation beyond this baseline; DOL does not prescribe that curriculum. The OECD and European Commission’s *Empowering Learners for the Age of AI* discusses related themes, including understanding systems, evaluating outputs, and ethical use. Its stated audience is primary and secondary education, however, so it is not an adult workplace standard or a specification for a particular job. The broad overlap is understanding systems and judging outputs. For an adult learner, the DOL framework is the relevant workplace guide. An explanation and checked sample can reveal the next gap, giving the learner a concrete question for a manager or instructor. A course badge alone does not prove competence.

Sources: Attachment I (Accessible PDF): The U.S. Department of Labor's Artificial Intelligence Literacy Framework; Empowering Learners for the Age of AI

How can one real task turn the framework into a learning loop?

Consider an example: a procurement analyst regularly prepares a short summary of vendor proposals for an internal review. That task may look like a convenient place to try AI because the source is written material and the output is a draft. Convenience alone is not enough. Before experimenting, the analyst should check whether the employer permits the tool and whether proposal material can be entered into it. If either answer is unclear, use a public or otherwise approved sample, or ask before proceeding. A low-risk exercise is only low-risk when its data and use are authorized. The U.S. Department of Labor’s Artificial Intelligence Literacy Framework suggests that workers identify routine tasks, experiment with an AI tool, and compare its output with their normal approach, noticing possible time savings and where human oversight matters. That is a starting exercise, not evidence that AI improves every task. To make it informative, first write down what a useful summary must preserve: for example, the proposal’s scope, price, delivery conditions, exceptions, and questions that require a human decision. These are acceptance criteria for this illustrative task, not claims about a universal procurement process. Next, establish a small baseline. Using a permitted, non-sensitive sample, note how long the ordinary summary takes and what checks are normally required. It is to have a fair reference point, so a fluent-looking draft does not get mistaken for a better result. If the ordinary process includes comparing a proposal with requirements or resolving ambiguous terms, keep those checks visible in the baseline rather than treating them as overhead that AI automatically removes. Then try the tool on the same permitted source material. Direct it with the task, intended audience, required fields, and a clear instruction to flag missing or ambiguous information rather than fill gaps. This is a practical way to learn whether context and constraints change the usefulness of an output. It does not require a special prompt formula. The worker should avoid adding information that policy does not allow them to share, and should not assume that a system’s privacy or retention terms meet workplace requirements. Evaluation should be against the source and the accepted work standard, not against how polished the response sounds. Check each stated amount, condition, deadline, and named party against the proposal. Look for omitted exceptions, unsupported conclusions, and language that turns uncertainty into certainty. If the summary will inform a decision, keep the decision itself with the person or process authorized to make it. The framework’s emphasis on evaluating outputs and using AI responsibly supports this kind of review; the detailed checklist here is an editorial application to the example. Record a few observations: time spent producing the draft, time spent checking it, omissions or errors found, whether the required fields were present, and whether the result was useful for the intended reader. These notes are a proposed learning exercise, not a validated measurement method. A faster first draft can still cost more overall if verification takes longer or if a consequential omission is hard to detect. Conversely, a draft that helps organize material may be useful even when it does not reduce total time, provided the human review remains practical and the task rules permit its use. Use four questions to decide whether to continue: Was the task authorized? Were the consequences of an error limited? Could the work be reversed before anyone relied on it? Could a person check the output against a dependable source? A “no” or “not sure” answer is a reason to narrow the exercise, seek approval, or stop. For instance, a proposal summary used only as a private first draft is different from an unchecked summary sent to a supplier or used to approve spending. The example illustrates a decision boundary; it does not establish that any particular workplace permits either use. The result may be “do not use AI for this task.” If the tool misses important conditions, produces errors that are difficult to catch, or shifts too much effort into review, that is useful evidence about fit. The worker can revise the instructions, limit the tool to a smaller subtask, or return to the existing process. One completed, documented comparison teaches more about task fit and verification needs than collecting prompt tips without applying them. Keep the accepted quality standard intact, and ask a manager or training provider what approved tools, data boundaries, and review responsibilities apply before moving from a practice sample to real organizational work.

Sources: Attachment I (Accessible PDF): The U.S. Department of Labor's Artificial Intelligence Literacy Framework

Why does DOL specify how learning should happen as well as what to learn?

The five content areas describe what AI literacy covers; the seven delivery principles describe how programs can teach it in context. A topic list alone may leave workers without practice, a link to their duties, or a way to assess progress. DOL presents the principles as flexible program-design guidance across industries, roles, education, and workforce settings, not seven more subjects every employee must memorize. Its February 13, 2026 release calls the framework a resource for designing literacy efforts, not a standardized course.

The first test is whether learning connects to work and includes experience. DOL’s experiential-learning principle asks programs to let learners apply ideas to relevant tasks and reflect on what happened. A worker can notice whether a task suits the tool, what context it needed, and how much checking the result required. This is evidence from one exercise, not proof that AI improves every workflow. Contextual integration matters for the same reason: practice tied to a role can surface constraints a generic demonstration misses. Providers should be able to name the practice task .

Access and starting points affect whether practice is possible. The framework notes that learners may need digital skills, a device or reliable broadband. Its cohort-specific examples recommend adjusting pace and references to workers’ experience, technology familiarity, or career stage. A learner can ask whether the program assumes tools and support they can access, and whether it starts at a level that permits useful practice. These design choices concern participation; they do not define a person’s ability to progress.

DOL’s principle on complementary human skills pairs AI practice with problem solving, communication, creativity, and domain expertise. Its examples include critically examining results, refining drafts for an audience, and using subject knowledge to assess output. Context helps a worker spot when a polished answer misses a requirement; communication helps explain the limitation. This does not mean human judgment always succeeds or that tools cannot do useful work. It means learning should consider how people apply and assess AI, not treat generated output alone as proof of learning.

Other principles address continued-learning pathways, enablers such as managers or trainers, and agility as work changes. An introductory activity can identify a gap and suggest a next level without settling every future need. DOL calls the framework a starting point and says it will evolve with stakeholder input, AI advances, and labor-market changes. It offers a flexible structure, not an industry-specific curriculum, universal sequence, or evidence that a particular program works.

When comparing training, look for a relevant practice task, feedback, support suited to your starting point, and a next step if a gap appears. A lecture explains concepts and a badge records completion, but neither alone shows whether you can apply learning to your duties. Ask what learners do and how their work is assessed. Prompt templates may teach basics; without practice and review, they may not meet your goal. Use DOL’s principles as questions about course design, not a scorecard certifying course quality.

Sources: Attachment I (Accessible PDF): The U.S. Department of Labor's Artificial Intelligence Literacy Framework

What can exposure evidence tell a worker about the purpose of literacy?

Exposure evidence directs attention to tasks that may change; it cannot tell a worker what an employer has adopted or whether that worker will be displaced. The International Labour Organization’s *Generative AI and Jobs: A Refined Global Index of Occupational Exposure* is a global, task-based analysis, not an individual risk assessment. It concludes that transformation is more likely than full occupational replacement because most occupations contain tasks requiring human input. That is a reason to inspect changing work, not a promise of preservation. The method explains the limit of that conclusion. The ILO combined task-level data, expert input, and model predictions. Its assessment drew on 29,753 tasks in Poland’s occupational classification, a survey of 1,640 employed people across ISCO-08 groups, and 52,558 data points on perceived automation potential for 2,861 tasks. Researchers compared these inputs with surveys and Delphi-style discussions among international experts, then used the resulting task knowledge to predict scores for ISCO-08 tasks. This supports occupational exposure estimates, not workplace observation, employer adoption counts, or performance on a particular organization’s data. The report estimates that one in four workers globally is in an occupation with some generative AI exposure, while 3.3 percent of global employment falls in its highest exposure category. These are global estimates across four gradients. They do not mean that a quarter of workers will lose work or that someone in the highest gradient has a particular chance of dismissal. Exposure concerns tasks that may be affected under the index’s method. Technical capability, observed use, employer adoption, duty redesign, labor demand, and displacement remain separate questions. Adoption may change tasks rather than headcount; review, accountability, sensitive information, or other constraints may also limit use. The index cannot resolve those local facts. That boundary makes literacy practical. The Department of Labor framework asks programs to help workers understand AI, explore uses, direct systems, evaluate outputs, and use AI responsibly. Applied to exposure evidence, these capabilities can help a worker notice changing task elements, check output against a work standard, and identify where judgment or escalation remains necessary. This link is editorial interpretation, not a finding that the ILO tested DOL’s framework or that literacy prevents job loss. For a recurring drafting task, a worker might separate producing a first draft from approving it, then ask what supports the output and who is accountable for the final decision. This does not imply every employer should use AI for drafting. It makes boundaries clear enough to ask about policy, checks, and responsibility. If no approved tool or safe practice data is available, mapping the task and its verification standard can still be useful. For a career decision, use the ILO analysis to ask which duties in your role are changing, what local employers require, and what your workplace permits. Do not infer adoption from exposure or treat the framework as protection from displacement. A task review can clarify a learning gap; local requirements and workplace rules must supply evidence that a global index cannot.

Sources: Generative AI and Jobs: A Refined Global Index of Occupational Exposure

Where does worker responsibility end and organizational risk ownership begin?

A worker's responsibility is to use an approved system within its stated limits, protect information, check outputs against the work standard, and raise a concern when the result or use looks unsafe. That responsibility does not make the worker the authority who approves a high-stakes AI deployment or the sole owner of its risks. The Department of Labor's AI literacy framework places responsible use among the competencies workers should learn, including awareness of sensitive information, workplace rules, context-specific risk, and accountability for decisions and outputs. The worker-level duties are to notice, verify, follow policy, and communicate; organizational governance remains with the organization.

NIST's AI Risk Management Framework helps clarify the difference in scale. Its Core organizes organizational risk work under four functions: Govern, Map, Measure, and Manage. Govern covers policies, documented roles, communication, and accountability structures; Map establishes context, intended use, potential impacts, and limits; Measure evaluates and monitors performance and risk; Manage prioritizes responses, allocates resources, and plans for incidents or recovery. NIST describes these as functions for managing AI risks across system lifecycles, not as a personal literacy syllabus. It explicitly says the actions are not a checklist and are not necessarily ordered steps. The framework is voluntary, and organizations can adapt it to their needs and capabilities. The functions offer vocabulary for asking who owns what, not a personal checklist.

A worker who knows a process well can contribute information that a system owner may not see from technical documentation alone: which cases are unusual, what a correct result must include, who will rely on it, and what a mistake could affect. In NIST's Map function, context and perspectives from domain experts, users, and potentially affected people help test assumptions about use and identify limitations or impacts. A worker might report an omitted condition, a recurring poor result, or use beyond the stated purpose. This is useful evidence for people responsible for evaluation and policy; it does not itself authorize the worker to broaden the system's use, set the organization's risk tolerance, or decide that the system is safe for deployment. NIST assigns organizational roles and lines of communication a place in Govern, including leadership responsibility for decisions about AI system development and deployment.

The level of caution should track the consequence and reversibility of the task. A draft for internal brainstorming that contains no restricted information may be easy to check and discard. A recommendation that could affect someone's money, benefits, safety, or access to a service calls for clearer authority, review standards, and a route to challenge or correct an error. This applies DOL's task-and-context emphasis and NIST's focus on organizational impact and oversight; neither sets a universal employer permission rule. A worker should therefore confirm the local rule rather than infer approval from a general framework. Before an experiment, ask which tools are approved, what data may be entered, who must review the result, and where errors or near misses should be reported. If those answers are unclear for a consequential task, pause and ask the manager or designated owner before proceeding. Recognizing that a use is outside one's authority is part of responsible literacy, not a failure to learn the tool.

Sources: Attachment I (Accessible PDF): The U.S. Department of Labor's Artificial Intelligence Literacy Framework; AI Risk Management Framework: Core

Which learning route fits the outcome, prerequisites, and constraints?

The right learning route depends on what you want to do with AI. If your goal is to improve work in your current role, start with a permitted task and a small amount of structured learning, then seek feedback on the result. If you want an adjacent AI-enabled role, add a project that demonstrates relevant work and compare it with real local job requirements. A degree becomes more proportionate when the target requires sustained technical foundations or a formal credential. Baseline AI literacy alone is not preparation for machine-learning engineering or research. That distinction fits the Department of Labor’s framework: it is a broad foundation for learning, not a prescribed course or credential. The OECD’s 2026 policy paper, *Skills in the AI Age*, likewise recommends both broad AI literacy and deeper formal education for developing technical talent. It also calls for flexible, modular online pathways and short, targeted adult reskilling. These are system-level recommendations about access and design; they do not show that a particular course or certificate improves one learner’s hiring chances. The useful inference is to choose depth by intended work, then check whether the learning produces evidence you can use. Self-study is flexible and often inexpensive, but may offer little feedback. A focused course or certificate adds structure, though content and assessment vary. A project or apprenticeship can show applied work and invite critique, but requires access, time, and guidance. A degree offers sustained study and assessment at greater cost and time; check whether your target actually requires one. For current-role use, begin with self-study or a focused course only if it helps you complete a supervised work sample. Document the task, checks, and points requiring human judgment; this is evidence of one bounded attempt, not broad proficiency or a career guarantee. Follow workplace rules on tools and data. If you are considering a move into an adjacent role, first inspect several current postings in your location or for roles you can realistically access. Separate recurring requirements from optional tool names. Choose a project testing a relevant skill; use a course to fill a gap and seek critique where possible. A project does not replace a required license, degree, or experience. A software-building goal calls for more than knowing how to use an AI assistant in an existing workflow. It may require programming, debugging, testing, and systems knowledge. Machine-learning engineering or research generally calls for deeper mathematics, statistics, computing, and data preparation. A degree can provide sustained, assessed foundations; a course or self-study may address narrower gaps. Compare curricula and prerequisites before enrolling. The DOL framework does not prescribe any of these routes, and its baseline is not a technical qualification. Make the choice against your constraints as well as your ambition. Set a budget and weekly study limit; account for caregiving, health, location, schedule, and required income. Start with a reversible step: one permitted task experiment, short contextual module, or small project with feedback. Define what you will demonstrate and what would justify continuing. Invest in a gap revealed by the work, not generalized anxiety. The OECD’s recommendations support flexible access to learning and distinguish broad literacy from advanced AI skills; they do not rank these four routes for an individual or promise employment outcomes. A practical sequence is therefore: test current-role use with one checked work sample; add structured instruction or feedback where the sample exposes a gap; pursue a larger credential when local role requirements and your target justify its time and cost. Before paying, compare the provider’s current curriculum, prerequisites, assessment, support, total cost, and schedule with the work you intend to do. No route by itself guarantees job readiness, employment, or income.

Sources: Skills in the AI Age; Attachment I (Accessible PDF): The U.S. Department of Labor's Artificial Intelligence Literacy Framework

What is the smallest useful next step?

Choose one recurring task and write down what a good result must contain, what information may be used, and who can approve the work. Practice on a permitted, low-consequence example. Compare the output with the usual quality standard and note what a person still has to verify. The U.S. Department of Labor’s AI Literacy Framework offers a starting point: it describes workers experimenting with routine tasks and comparing AI output with their usual approach. That example is guidance, not proof that every task is suitable or that AI will save time. Follow workplace rules and ask before using real organizational information or an unapproved tool. If review ownership is unclear for a consequential result, pause and clarify responsibility. This test can show whether the task is a sensible place to learn, where review effort appears, and which skill gap deserves attention. It cannot establish what an employer will adopt or whether a role will change. Your goal and constraints determine whether to practice more or study deeper. Ask a manager or provider: “What capability should I demonstrate next, and what feedback or approved practice can help?” The free task-level change-pressure checker can organize questions; its signals are not validated probabilities of job loss and do not replace workplace rules.

Sources: Attachment I (Accessible PDF): The U.S. Department of Labor's Artificial Intelligence Literacy Framework

Questions readers ask