Humans can genuinely recognize that something is important and then act as if they do not: discuss diets while munching on candy, or complete security training only to fall victim to phishing when distracted.
Now that AI ethics in the workplace and AI literacy are hot topics, this problem resurfaces. Enterprise training needs not just to inform, but help turn broad principles into everyday habits.
Meanwhile, people are using AI at work with or without training or regulations. As of early 2025, research by the University of Melbourne and KPMG International found that among nearly 48,000 respondents across 47 countries, 58 percent intentionally used AI at work. Of these users, 56 percent admitted making mistakes because of AI, and 57 percent concealed AI use from employers (N. Gillespie. Trust, attitudes and use of artificial intelligence. kpmg. 2025).
Now that organizations have moved from whether to let people use AI to how to let them do so, it is time to think of what a successful AI ethics training program should look like.
In this article, we draw on our experience at Blue Carrot designing various courses oriented at skill development and internalization, not just theoretical knowledge — and apply them to AI ethics. These insights are meant to help not just define a curriculum but implement it while securing a lasting effect.🧐
Summary
- Key takeaways
- What is AI ethics training for employees?
- Why AI ethics training matters now
- Relevant AI regulations
- What should an AI ethics training program include?
- Designing AI ethics training: from information to behavior change
- Best formats for AI ethics training
- How to build and roll out an AI ethics training program
- Common challenges (when just a course is not enough)
- AI ethics program rollout checklist
- How to measure AI ethics training success
- Conclusion
Key takeaways
- AI ethics training is not just for technical teams because AI-based tools are used across different roles.
- An AI ethics program needs both a shared foundation and role-specific competencies to succeed.
- The goal is behavior change, not just awareness. Scenario-based learning, practice, and ongoing reinforcement are more likely to achieve this than one-off modules.
- Regulations set the overall direction, but training should be focused on real work. Memorizing compliance requirements is not the optimal way to teach; fitting them onto daily workflows is.
- Successful AI ethics initiatives should combine content with instructional design and meaningful metrics.
What is AI ethics training for employees?
AI ethics training for employees is workplace training that teaches people to use AI responsibly and safely, in line with organizational and compliance requirements. Its goal is not only to build awareness, but also to develop practical habits for daily work with AI.
In practice, AI ethics training does not only focus on ethics in the philosophical sense. There is an ethical component to most of the principles of responsible AI use, so a good program will not just focus on “is it morally right?” but also on “here is a common trap” questions.
For example, in the research above, 48 percent admitted uploading company data into public AI tools (N. Gillespie. Trust, attitudes and use of artificial intelligence. kpmg. 2025). This can be viewed as both an ethics and security problem. The task of a course in this scenario is not to disentangle these aspects, but to prevent this from becoming a risk (and to demonstrate when and why it becomes wrong).
Why AI ethics training matters now
Several distinct factors make AI ethics and literacy training a pressing concern — not only prompting rapid adoption, but also encouraging competition around whose program is better:
- Employees find and use AI tools on their own. Importantly, one does not need to understand AI to use it, which means risks grow.
- Regulators and clients now start to expect responsible AI practices. For example, the European Commission has even created a repository of 40+ AI literacy practices collected from organizations — reflecting the trend (AI Literacy Practices. European Commission Logo. 2026). Similarly, “we use AI responsibly” is a stronger customer-facing claim.
- After risks have become quantified, responsible AI use has become a business capability in its own right. A 2025 analysis by The Conference Board and ESGAUGE found that 72 percent of S&P 500 companies disclosed at least one material AI risk in their annual filings — compared to 12 percent in 2023 (Matteo Tonello. AI Risk Disclosures in the S&P 500: Reputation, Cybersecurity, and Regulation. The Harvard Law School Forum on Corporate Governance. 2025). This is directly linked to AI ethics, either through corporate responsibility or lawsuit risks.
Relevant AI regulations
AI regulations are relatively new, but they provide an important bridge between the corporate best practices that defend the company and its clients — and the measures that protect the wider business and social landscape.
Here are the most commonly referenced ones:
- European Union AI Act — contains a risk-based framework for AI systems, designed in 2024, with most rules applicable since August 2026. Article 4 requires providers and “deployers” (i.e., user organizations) of AI systems to ensure AI literacy among their staff. It does not prescribe a standardized training course, though.
- NIST AI Risk Management Framework (AI RMF) — released in 2023 and used in the United States in the absence of a single federal law. The framework is voluntary and meant to provide guidance. More regulations are emerging at the state level, like Colorado’s AI Act, etc.
- Industry-specific requirements — AI use is a new field where existing sector-specific compliance rules apply in new ways. For example, American healthcare organizations using AI to process protected health information still need to account for HIPAA. Similarly, organizations using AI in employment decisions need to consider applicable anti-discrimination and employment laws, etc.
- International standards — these matter in terms of business function and contractability. For example, ISO/IEC 42001:2023 (the international standard for AI management systems) does not prescribe how particular AI models should work, but covers organizational areas like AI policies, defined responsibilities, etc.
An off-the-shelf AI ethics course will reflect a particular combination of assumptions about which regulations and standards apply — meaning it is nearly impossible to mechanically transplant a course into a different business environment.
And yet, there are general best practices that govern the ethical use of AI in any situation — they just mean different behaviors and routines in different contexts.
What should an AI ethics training program include?
All this means a practical AI ethics training program will never be quite the same everywhere. It will have a “generic” part with must-have topics for everyone, and a role-specific and industry-specific part.
👉 The shared foundation: must-have topics
Regardless of role or industry, there are topics relevant to anyone working with AI, and a common understanding of what responsible use of AI even means is essential. Such a shared foundation typically covers:
- AI literacy and limitations: how AI systems generally work, what they can and cannot reliably do, and why they are not the “ultimate source of truth”;
- Privacy and data protection: what information can be “given” to AI tools, which tools are approved, and how to work with sensitive, personal, confidential, or proprietary data;
- Security: common AI-related security risks, like unauthorized tools, data leakage, malicious or manipulated inputs, and unsafe integrations;
- Bias, fairness, discrimination: how AI systems can reproduce or amplify biases and what potentially unfair outcomes this can lead to;
- Human supervision and accountability: what situations need human review and verification (when to escalate decisions to human colleagues);
- AI transparency and responsible use: when to disclose AI use, how to distinguish information from generative AI from verified sources, and what uses are acceptable by organizational policies;
- Intellectual property and content ownership: what risks emerge when using copyrighted, confidential, or third-party materials with AI systems;
- Corporate policies and escalation: what is approved, what is to document, where to escalate when a potentially harmful situation arises.
The goal here is not for everyone on board to be an “AI guru” but simply to ensure people are knowledgeable enough to recognize common ethical risks and act properly.
👉 Role-specific learning paths
The shared foundation can then branch according to how employees actually use AI. Different roles will experience the same general AI-related ethical problems in different kinds of situations and will need different practices.
Such role-specific corporate training can include very diverse topics:
- HR and recruitment: bias and discrimination, candidate privacy, transparency, and oversight tactics when it comes to employment decisions;
- Finance: model risk, fairness, explainability, sensitive financial information;
- Managers and decision-makers: AI governance, risk assessment, approving use cases, accountability;
- Developers and technical teams: secure AI integration, testing and monitoring, data control, model limitations, documentation, responsible development practices.
Designing AI ethics training: from information to behavior change
There is always the temptation to treat such courses as retellings of the corresponding policies, paragraph by paragraph, but in a graphically “fancy” format (which, sadly, often boils down to video → quiz → video → quiz).
However, this approach fails to achieve the main objective: enabling the learner to recognize the relevant situation, make the right decision, and understand why. In ethics specifically, real-world decisions are contextual.
For example, it is all too easy to remember “do not upload sensitive information” as a mantra. However, the actual skill is to automatically see that a superficially innocent prompt contains such information.
Here are some of the most important aspects of turning raw information into an actual training program, based on Blue Carrot’s experience with various corporate learning projects.
📌 Turning policies and regulations into practical cases
Policies and regulations are primarily documents — conceptually dense, packing as much information as possible in rigid formulations and paragraphs. This format is meant for consulting and looking up, not for learning new information. When faced with a document-based course, a learner must figure out multiple things simultaneously. The main mental effort is spent on making sense of what is meant, not on applying it to real cases.
A more productive approach will introduce simple cases first, then raise complexity. For instance, instead of:
Employees must not enter personal or confidential data into unauthorized AI systems.
A well-designed course will present a case:
You are preparing a customer report and want an AI assistant to summarize it. It contains customer names, contact details, and transaction history. The AI tool is not on the company’s approved list. What do you do? (plus options)
In this way, the learner is not memorizing “privacy = good”; they are practicing a decision.
This is in line with the cognitive load theory used in pedagogy. In a recent revision of this theory, Chen, Paas, and Sweller (Ouhao Chen, Fred Paas , John Sweller. A Cognitive Load Theory Approach to Defining and Measuring Task Complexity Through Element Interactivity – Educational Psychology Review. SpringerLink. 2023) emphasize that learning difficulty depends not simply on the amount of information presented, but on how many interacting elements learners need to process and coordinate at once. For complex subjects like AI ethics for employees, this means that the right approach is to build connections between concepts gradually — the opposite of what documented policies usually look like.
📌 Balancing compliance requirements with engaging learning
“Engaging” does not necessarily mean “entertaining.” Turning the EU AI Act into a TV series would ultimately just produce an incredibly stale TV series, not endlessly “cool” learning content. Instead, engagement comes from:
- Relevance and relatability;
- Meaningful choices and consequences;
- Context (needed to instigate curiosity).
This challenge is now efficiently solved with storytelling in e-learning — a method that has proven useful across dozens of projects in Blue Carrot’s practice, especially with seemingly abstract topics. It works well because a story puts the rule into a relatable context:
An employee uses an AI-generated summary for a client meeting, then discovers an invented/hallucinated figure, and has to decide what to do about it.
Storytelling is not the only way to improve engagement — but for courses like AI literacy and ethics, it works better than trying to gamify the learning process or using jokes in the content at every opportunity.
📌 Using instructional design to improve retention
Finally, everything needs to be cemented together with proper instructional design — that is, the set of principles that makes the material learnable. These principles include:
- Identifying learning objectives before settling on content. Instead of asking, “what should we tell?”, ask “what should they be able to do?” — this prevents the course from becoming a policy retelling in disguise.
- Providing opportunities to retrieve and practice the insights: scenario questions, branching decisions, properly spaced refreshers.
- Ensuring feedback also teaches and explains something — not just “Wrong. Try again” but also why.
- Progressing from simple to complex. Real ethical decisions rarely come labeled as “privacy question”, etc., but for the sake of training, let the situations progress from straightforward to realistic.
- Reinforcing skills after the course, from microlearning to periodic scenarios and established channels for updates.
In other words, if the desired behavior matters six months after completion, the learning experience should not disappear six minutes after completion. Our instructional design services focus specifically on ensuring higher knowledge retention, since we often deal with compliance training.
The training then moves from information through practical application to actual behavior changes outside of the training environment.
Best formats for AI ethics training
The best format for AI ethics training depends on what employees need to learn. For a shared foundation, a short self-paced module may be enough; for practical skills (which are typically role-specific), more interaction is needed.
- Self-paced e-learning: interactive online modules for basic literacy, core policies, and the like — easily scalable and non-disruptive;
- Scenario- and simulation-based learning: these allow one to progress beyond memorization with modeling situations close to actual work;
- Blended e-learning solutions and VILT: a combination of self-paced modules with virtual instructor-led training, as well as workshops, is arguably the most optimal approach: a common base with enough space for ambiguous cases;
- Rapid e-learning: ideal for changing regulations or realities because it allows the deployment of focused learning interventions without a lengthy development cycle.
In practice, the important question is not which format is the most spectacular by itself, but which fits the actual needs of the course.
How to build and roll out an AI ethics training program
Building an entire program from scratch may look intimidating, but in reality, following a structured approach helps make the process predictable and more easily manageable.
👉 #1 Assess your organization’s AI usage and risks
Start with what is actually happening rather than what the policy says should be happening. Work with IT, security, legal, and other teams to identify which AI tools employees use, when, and with what data, to understand the actual risks.
👉 #2 Define training goals, audiences, and success criteria
Together with HR/L&D and business leaders, decide what different groups should be able to do after training and how to know when they are. Separate organization-wide goals from role-specific ones and establish the metrics. Now, a learning design partner can step in and help translate the requirements into measurable learning objectives.
👉 #3 Align training content with policies and requirements
Have legal, compliance, security, and subject-matter experts validate the relevant details (requirements, tools, procedures). A training provider or instructional-design team can then convert this material into learner-facing content and scenarios.
👉 #4 Launch the program and encourage employee adoption
Treat the rollout as a change management exercise in which HR/L&D and managers explain why the training matters and organize the process.
👉 #5 Reinforce learning through updates and continuous improvement
L&D and program owners can monitor feedback, reported incidents, policy changes, and observed behavior — these signals are useful to define what needs updating or when refreshers are due.
Common challenges (when just a course is not enough)
Sometimes, even a perfectly produced course can still fail because of the wrong conditions around it. Employees may pass the quizzes without connecting them to their actual workflows, the policy may be out of sync with the tools in use, etc.
The fix is partly instructional: realistic scenarios (based on actual cases, not assumptions), explaining all the reasoning, etc. It is also operational: keeping policies and approved tool lists up to date, providing very clear escalation routes, and using incidents to update training.
There is also the human factor: employees feel it when a course disrupts their work. If engagement is not what it should be, instead of sending more reminders, diagnose the problem: is it the course, the workload, the format, or something else? This information is useful in perfecting the course after the initial launch.
AI ethics program rollout checklist
Based on our practice, several crucial questions need to be answered at some point during the project — and preferably early on, to avoid confusion. It makes sense to treat them as a checklist:
- Have we identified which AI tools and use cases employees actually use, including unofficial ones?
- Have IT, security, legal/compliance, and relevant business teams identified the exact situations employees need to handle correctly?
- Does each role receive the guidance it needs, rather than another generic AI course?
- Are the exercises similar enough to employees’ actual routines?
- Are approved tools, policies, escalation routes, and responsible contacts clear and accessible?
- Does the training ask employees to recite rules or apply principles to ambiguous situations?
- Can employees complete the training without disrupting their work, and do they understand why it matters?
- Who owns updates once they are needed?
- How will we know whether employees actually apply what they learned?
These questions are the absolute foundation. For a more detailed strategy, you can download our AI Ethics Training Rollout Checklist, which provides practical guidance for before, during, and after rollout.
How to measure AI ethics training success
While completion rates and quiz scores are useful, they only tell part of the story. An employee can complete every module and still make the wrong decision when a real situation arises. Accordingly, the metrics need to be quite diverse.
- First of all, track participation and knowledge: completion rates, assessment scores, time spent on modules, etc. This allows you to pinpoint some issues with the training format.
- Then, monitor the metrics that show how the learnings are applied, e.g., use of approved vs. unapproved AI tools, number and quality of AI-related escalation requests, etc.
- Finally, assess organizational outcomes: incidents involving unauthorized AI use, data exposure, or policy violations, compliance with approved workflows; or reductions in recurring errors (from audit data).
Not every organization will have all the data needed, so it makes sense to choose a small set of meaningful indicators before launch and compare them over time.
Conclusion
AI ethics training could theoretically be about turning everyone into an AI expert — but it should not. It could also be (and often is) just another box to tick. But in reality, it is about tailoring the general principles to the very concrete realities.
This means accurate content is not enough: to really work, this kind of course needs instructional design, realistic scenarios, carefully chosen formats, and reinforcement tactics.
At Blue Carrot, we combine instructional design expertise with experience creating engaging e-learning for complex subjects. Whether you need a shared AI literacy foundation, role-specific training, or a combination of both, we can help turn your requirements into learning experiences that drive real-world behavior change.
Looking to build an AI ethics training program that employees will actually use? Talk to Blue Carrot about a custom solution for your organization.
