"We paid for AI training this year. Why can nobody tell me whether it worked?" That is the question we hear most often from owners and department heads once the first round of courses is finished, and it usually has the same answer: the program was never built against a standard, so there is nothing to check it against.
The U.S. Department of Labor fixed the missing-standard part on February 13, 2026, when its Employment and Training Administration issued Training and Employment Notice 07-25, the federal AI Literacy Framework. What it did not do is hand employers a working AI literacy training checklist. This article turns the framework into one you can use to build a program or audit the one you already have.
It is a set of pass-or-fail checks drawn from the framework's five content areas and seven delivery principles, applied to your actual roles and tools. Instead of asking "did people attend," it asks whether each person practiced on real tasks, can judge AI output, has a manager who reinforces the skill, and whether the content gets updated as tools change.
The framework itself is voluntary. The department's full TEN 07-25 document describes it as guidance "to support the evaluation of high-quality AI literacy content," and SHRM's coverage of the release notes that it imposes no new requirements on companies or workers. That is exactly why it works as an audit tool: it gives you a neutral, published yardstick that no training vendor wrote.
If you want the plain-English tour of what the framework contains, our earlier walkthrough of the DOL's AI literacy standard covers the five content areas one by one. This piece assumes you know the structure and want to put it to work.
Here is the take most teams resist: the five content areas are the easy half. Almost any course catalog can claim to "cover" understanding AI, exploring uses, prompting, evaluating output, and responsible use. Topic coverage is cheap to promise and hard to disprove, so auditing a program by its syllabus nearly always produces a passing grade.
The seven delivery principles are where programs actually fail, because they describe how learning has to happen, and you can observe whether it did. The framework's first principle is blunt: AI literacy "is most effectively developed through direct, hands-on use," and workers build understanding "not by reading about AI in the abstract." A program built on recorded videos and a quiz can cover all five topics and still miss most of the seven principles.
The research lines up with that. In a DataCamp survey of more than 500 U.S. and UK enterprise leaders conducted with YouGov, 82 percent of organizations offered some form of AI training, yet 59 percent still reported an AI skills gap, and video-based courses were the most common format. On the employee side, Docebo's AI Readiness Gap report found that 85 percent of employees say the training they receive does not help them use AI in their role. Those are delivery failures, not content failures, so the checklist below weights delivery heavily.
A checklist is only as honest as the scope you apply it to. Before scoring, list every role that touches an AI tool, which tools each role uses, and what decisions or outputs leave that role with AI involved. A payroll specialist drafting employee communications in Copilot and a sales lead summarizing calls in a meeting assistant need different evidence, even though both sit under "AI literacy."
Then pick one scoring rule and stick to it. We recommend a simple three-point scale per item: 0 if there is no evidence, 1 if it exists for some roles or only on paper, and 2 if you can show it working for every role in scope. Scoring by evidence, not by intent, keeps the exercise from turning into a self-assessment of good intentions. It also gives you a baseline you can rescore in six months, which is the only honest way to measure AI training ROI without inventing the numbers.
The framework's own descriptions are more specific than most summaries suggest, and that specificity is what makes them auditable. The table maps each content area to the sub-skills the department lists and to the evidence that shows a person has them, rather than that they sat through a module about them.
| Content area | What the framework includes | Evidence to look for |
|---|---|---|
| Understand AI Principles | Probabilistic outputs, hallucinations and accuracy limits, human design and oversight | Staff can explain, for their own tool, why it can be confidently wrong |
| Explore AI Uses | Productivity, information support, task-specific and decision-support uses | A role-level list of approved and off-limits tasks that people can recite |
| Direct AI Effectively | Context, clear prompts, relevant input data, iteration | Saved prompts or templates for recurring tasks in that role |
| Evaluate AI Outputs | Accuracy, completeness, logical errors, alignment with intent, need for human judgment | A documented review step before AI output reaches a client, record, or decision |
| Use AI Responsibly | Protecting sensitive information, following workplace policy, maintaining accountability | Staff can name what data must never go into which tool, and who is accountable |
Score each row per role. The two rows that most often come back as zero are evaluation and responsible use, because they require a person to apply judgment inside their own workflow, and that cannot be demonstrated in a generic course. If your team cannot name the data that stays out of AI tools, a short exercise built on a concrete never-paste list usually closes that gap faster than another module.
These are the items to score hardest, because each principle in the framework comes with concrete examples you can look for. Work through them in order; the first three carry most of the weight for a 20 to 300 person company.
Item three deserves extra attention. In most environments we review, managers are trained alongside their teams or not at all, so nobody is equipped to catch weak AI-assisted work during normal review. The difference between AI training for employees and AI training for leaders is exactly what this principle is asking you to plan for. It is also the leadership behavior we describe in leadership's role in building cyber awareness: people take a skill seriously when their manager visibly checks for it.
A passing program scores 2 on every content row and on the first three delivery items for every role in scope, and at least 1 on the remaining four. Anything below that is a program with gaps you can name, which is more useful than a vague sense that training "went fine."
Do not chase a perfect score on the first pass. The point of the checklist is to show leadership where the next dollar of training budget should go. In our experience, that is almost always toward hands-on, role-specific practice and manager preparation, not toward more topics. That matches the argument in why small teams need practical AI training rather than more tools: the gap is applied skill, and applied skill only comes from practice on real work.
One named person should own it, and it should not be the vendor that sells you the training. In a 50-person company that is usually an operations leader or HR lead with authority to change the program, supported by whoever administers your AI tools so the tool inventory stays accurate.
The owner's job is to rescore on a schedule and whenever something material changes: a new AI feature switched on across Microsoft 365, a new department adopting a tool, or a policy update. That ties the agility principle to a real trigger instead of a calendar reminder nobody honors. It is also one of the skills we flagged in the AI skills every SMB leader should cultivate: knowing how to tell whether your team's capability is actually improving.
Keep the evidence with the checklist. Exercise outputs, review steps, manager sign-offs, and update logs are the artifacts that prove the program is running. The same records that answer "did it work" also happen to be the documentation regulators and insurers increasingly ask for, which turns a training exercise into a governance asset at no extra cost.
When we run this audit with a client, we start from their tool inventory and access logs rather than a survey, so the scope reflects the AI people actually use, including features that arrived switched on inside software they already pay for. From there we score each role against the content and delivery checks, show leadership the two or three gaps that matter most, and build role-specific exercises from the client's own documents and workflows.
Because we also manage the environments those tools run in, we can tie the responsible-use checks to real controls, such as which data labels block AI access and which tools are sanctioned. That keeps the training honest: people practice inside the same guardrails they work under every day, and the checklist evidence lines up with what the systems actually enforce.
Run the checklist once on your highest-use department before you buy another course. The scores will tell you whether you need more topics or, far more likely, more hands-on practice and better-prepared managers.
If your team is moving faster with AI than your guardrails are, start with structured training rather than another tool. AI University gives your people role-based AI training with security built into the material, not bolted on afterward.
If you would rather talk through your specific environment first, book a strategy call with Securafy and we will walk your current AI usage, exposure, and the fastest path to safe adoption.