Before the widespread adoption of modern automated platforms, data governance programs often included training as a part of change management activities, alongside various forms of support in building a “data citizenship” culture within the organization.
Typically, employees completed an annual computer-based training that emphasized the importance of compliance, or received an invitation to an awareness campaign – often ignored, as there was no direct connection between “data governance” and their daily tasks. Everyone agreed that building data culture was a slow, gradual process and that even if they missed training, there was always time to attend it next quarter.
The introduction of generative AI, and in particular, AI assistants, into everyday working life has rendered this lack of data governance training problematic for organizations. Artificial intelligence systems not only introduce risks that organizations have never mitigated, but they also increase the probability of immediate adverse consequences in a completely different way from historical data compliance breaches.
Yet, many organizations still treat AI literacy as an afterthought, leaving their workforce between equally damaging extremes: ungoverned experimentation and avoidance.
Ungoverned experimentation is not a part of creative development or a brainstorming activity performed in a safe environment. It is a situation where employees embrace AI systems – generative AI, in particular – under the misapprehension that enterprise license terms, or their own access to data, mean that they can give AI assistants the same data they are working from. This trust in the intrinsic confidentiality of an AI assistant is prevalent in many organizations. When it is combined with a lack of knowing how to formulate a correct prompt and how to check whether the output is actually true, the risk of a decision being made on incorrect AI output increases dramatically.
On the other hand, some organizations are so terrified of legal liability, intellectual property exposure, or damage to their reputation that they ban all AI tools by default. This seemingly easy solution brings more problems than ungoverned experimentation – in particular, the rise of shadow AI. Similar to shadow IT, shadow AI is the workforce’s use of AI systems that haven’t been approved for use by leadership, IT, legal, and risk compliance teams. Such use still has the same consequences as ungoverned experimentation: risk of data leak, risk of compliance and regulatory violations, and poor traceability.
In order to not fall foul of these two extremes, an organization must have a user enablement activity as a part of an AI governance implementation or AI adoption project rollout.
The fundamental part of a user enablement activity is its three-tier educational structure. Training material, examples, and tools are structured according to the employee’s operational role and broadly grouped into three main tiers: leadership and senior management, governance and AI champions, and end users. Depending on whether the organization is a producer or deployer of AI systems, it might add training for technical teams as a fourth tier.
Tier 1: Leadership and Senior Management Education
Executives are responsible for AI strategic direction, scope, and intent of AI adoption within the organization and the distribution of resources. They also ultimately carry full legal responsibility for the organization. Thus, their AI education should skip over operational aspects of AI adoption and concentrate on the governance of oversight and strategic risk. It should explore:
- How the competitive necessity of AI adoption is balanced against financial, operational, and reputational risk
- How to navigate compliance mandates introduced by AI legislation (as an example, Article 4 of the EU AI Act that explicitly mandates AI literacy within the organization) and what is changing in the accountability of C-suite personnel
- How to establish a new AI aspect within existing data governance and risk management frameworks
- How to define new facets of emergency response protocols across the organization
Tier 2: Governance Teams and AI Champions / AI Power Users Education
Those who hold governance roles within the business, as well as those who act as “human-in-the-loop” for AI systems within the organization, need training focused on operational aspects of AI adoption. They must understand internal AI policy, as well as external government legislation (if it exists). They should also know their role in the risk assessment, vendor evaluation, and tool selection processes. Their education should cover:
- How to translate internal AI strategy or policy into day-to-day processes
- How to conduct risk assessments
- How to select the right vendor or the right tool
- How to monitor a deployed AI system, and what the mechanisms are for reporting hallucinations, bias, or other faults
- How to coach and mentor colleagues, and how to drive innovation within the department in a safe manner
Tier 3: End Users Education
AI education of the workforce is one of the pillars of the EU AI Act, the reason being that many AI systems are designed to be used by a wide population. As a consequence, it is important that this population knows how to use AI systems in a responsible manner. The education for this tier is concise, specific, bite-sized, and, ideally, integrated into the daily routine. It should cover:
- What data is allowed and what is not allowed to be used with AI systems
- How to critique AI output and how to develop healthy skepticism toward AI output
- How to create prompts that give as reliable results as possible without inadvertently exposing sensitive data
- How to recognize abnormal behavior, hallucinations, or bias in AI systems
- How to report, and if needed escalate, problems with AI systems or accidental data exposure
Real-World Examples
This approach is proving to be most optimal for several global companies.
For example, JPMorgan, Unilever, and Amazon, to name a few, are reported to have created a structured, role-segmented AI upskilling and compliance framework across their global workforce across similar tiers that we discussed here:
- Leaders/executives, sometimes functional leads
- Compliance, risk, engineering, AI champions
- General workforce
This approach of tailoring the depth, regulatory focus and practical skills of the training program is taking companies further than mandatory e-training modules of the past – it is building the competency that will become a competitive advantage of the future.
Legislation
While the U.S. currently doesn’t have a federal stature mandating AI literacy, the U.S. Department of labor issued Training and Employment Notice No.07-25, establishing a national framework for AI literacy programs. It suggests that AI literacy covers five core domains: technical understanding, critical evaluation, bias mitigation, privacy protocols, and safe prompting.
In addition, many U.S. states have established AI literacy guidance. The state of Colorado, for example, mandates deployers of high-risk AI systems to ensure that personnel who operate them are trained to manage algorithmic bias, perform risk assessments, and handle disclosures ().
The EU AI Act doesn’t specify the framework or approach but mandates providers and deployers of AI systems to “take measures to support the development of AI literacy” – specifically advising to consider the technical, educational, and level of experience of the personnel.
UNESCO published three AI competency frameworks:
- For students
- For teachers
- For civil servants
They set international standards for user-level critical thinking, data privacy concerns, and ethical and responsible AI interactions.
Operational Enablers
This three-tier role-based foundation for user enablement activity should be enriched by these operational mechanisms:
- Various training delivery models: Depending on the audience and the culture of the organization, training is delivered in various ways: informal lunch-and-learn sessions, videos, conversations with external experts, testimonials from data owners, learning sharing within the department, podcasts, quests, and games.
- Continuous feedback loop: The training program cannot be a one-way street. Two-way feedback channels must be established: practical sessions during office hours, a prompt repository for the enterprise, discussion of emerging use cases, and showcasing of successful solutions.
- Event-driven updates: AI capabilities evolve rapidly; therefore, refresher training should be connected directly to major model updates, the introduction of new AI tools, changes in strategy, intent, or scope of AI adoption, or external changes in regulatory matters or legislation.
- Outcome-based metrics: The success of the training is measured not by attendance or completion rates, but by monitoring practical changes, such as the reduction of shadow AI, the quality of incident reporting, and the quality and frequency of risk assessments.
Ultimately, building AI literacy as a valuable competency for the workforce includes moving away from the traditional one-size-fits-all corporate e-learning model toward role-specific tailored training where leadership, AI champions, and end users receive targeted information that allows them to use AI systems in a responsible and safe manner.
Tailored to the culture of the organization and supported by continuous feedback mechanisms, targeted updates, and appropriate metrics, three-tiered AI governance education provides organizations with true competency, allowing them to unlock the full value of artificial intelligence.
Build your AI governance skills in 2026.
Explore training programs on AI governance, data governance, and compliance for data practitioners.
