AI in Business: Shadow AI, AI Washing, Skill Erosion.
Shadow AI, AI Washing and Skill Erosion: Three realities to understand
There is a comfortable version of the AI-in-business story. One where Shadow AI, AI Washing, and Skill Erosion simply don’t exist. It goes something like this: tools are accessible, teams gradually get on board, and the organisation moves forward. No grand plan, but no chaos either. An organic, reasonable, pragmatic adoption.
This version is widespread in Switzerland. It is also, in many cases, inaccurate.
What researchers and analysts have been documenting for two years is clear: many companies that think they are integrating AI are in fact being swept along by it. And the most problematic effects are not immediately visible.
Shadow AI: When AI moves faster than the organisation
The first concept to understand is that of Shadow AI: phantom AI. It refers to the use of AI tools within an organisation without validation, supervision, or formal governance from management or IT teams.
This is not a marginal phenomenon. According to an IBM survey of 3,000 workers, more than one in three employees admit to having shared sensitive professional information with AI tools without their employer’s consent. A Menlo Security study reveals that 68% of employees use personal accounts to access tools like ChatGPT at work, and 57% of them do so with sensitive data.
Shadow AI rarely arises from bad intent. It arises from a gap: between what teams need to work effectively and what the organisation officially provides. When that gap is too wide, people find their own solutions. It is human nature. It is also, according to Gartner, one of the most underestimated strategic risks of 2026 — 49% of organisations expect to experience a Shadow AI-related incident within the next twelve months.
The risk is not only about security, even though the leakage of confidential data to external servers is real. It is also a risk to operational coherence: when each department uses its own tools with its own practices and no shared language, individual gains do not translate into collective performance.
AI Washing: The strategy that isn’t one
The second phenomenon is more uncomfortable to name, because it concerns not teams but leadership.
AI Washing (by analogy with greenwashing) refers to overstating or distorting the reality of one’s AI use. This can be external: displaying an “AI strategy” in communications without actual practices to back it up. It can also be internal: convincing oneself that you are “doing AI” because a few employees use ChatGPT, without governance, without training, without any reflection on actual use cases.
This second type is perhaps the most common, and the least documented. Compliance Week puts it this way: AI Washing “thrives in a climate where technological optimism is high, understanding is low, and oversight is far behind innovation.”
The issue is not moral. It is strategic. An organisation that believes it is more advanced than it really is does not allocate its resources correctly. It does not train its teams on what they actually need. It does not ask the right questions. And it often discovers the gap too late.
Skill Erosion: What nobody is measuring yet
This is the least visible of the three concepts, and the one with the most serious medium-term implications.
Skill Erosion refers to what happens when employees stop exercising certain cognitive abilities because they have delegated them to AI. In the short term, it is a productivity gain. Over time, it is a loss of mastery.
Gartner introduced the concept of “AI lock-in” to describe this mechanism: when employees hand off fundamental tasks to automation, their ability to question, interpret, or correct AI outputs gradually weakens. The organisation becomes dependent on tools it no longer truly understands. Gartner estimates that half of companies could face irreversible skills shortages by 2030 if this trend is not managed.
A concrete example: a communications professional who uses AI to produce all their content saves time. That is real and measurable. But if they stop exercising their own judgement alongside the tool, something is lost. The ability to build an original angle. To sense the right tone. To write under pressure when the tool falls short. This is not visible in short-term metrics. It becomes visible when a situation demands it.
The same mechanism can be observed in recruitment: an HR manager who systematically relies on AI for pre-selection may gradually lose the intuition that allowed them to identify atypical profiles — precisely the ones algorithms eliminate first.
Skill Erosion is particularly difficult to detect because it is asymptomatic: it advances without triggering an immediate alert, until a complex situation or an AI failure reveals that the backup human expertise is no longer there.
How do you stay in control of AI?
Faced with these three dynamics, there is no miracle solution. But there is an approach that fundamentally changes the posture, what organisational researchers call Human-AI Teaming: not managing AI, but building genuine collaboration between human teams and tools.
Human-AI Teaming is not just another item on a list of buzzwords. It is a structured way of thinking about the relationship between human capabilities and tool capabilities, not as substitution, but as active complementarity. According to the CHAI-T framework, it rests on five dimensions: information exchange, mutual learning, cross-validation, feedback, and reciprocal capability augmentation.
What concretely distinguishes an organisation practising Human-AI Teaming from one caught in Shadow AI: in the former, there is an explicit reflection on what skills humans must retain, and what AI can take on without weakening the whole. In the latter, that reflection has never taken place.
This is not a technical project. It is an organisational and managerial project. And that is precisely why it is so often put off.
Four questions to assess where you stand
There is no universal diagnostic. But these questions allow for an honest reflection on where an organisation truly stands.
- Shadow AI: If you asked your teams to list all the AI tools they used this week, would you be surprised? Most leadership teams would be. This is not a trust problem; it is a question of the gap between what the organisation offers and what employees need to work effectively. That gap deserves to be understood before it is addressed.
- AI Washing: Does your organisation have a formalised “AI strategy”? If so, do the people who work alongside you day-to-day know it and recognise it in their actual work? A strategy that exists on paper but not in practice is not yet a strategy; it is an intention. The distance between the two is often greater than one assumes.
- Skill Erosion: Are there skills your teams exercised regularly two years ago that they exercise less today because a tool handles them? If so, was that a deliberate choice or an unnoticed drift? The question is not about resisting automation, but about consciously deciding what human expertise to retain and why.
- Human-AI Teaming: When an important decision is made in your organisation based on an AI tool’s output, who validates it? Who is in a position to say “this result seems right” or “something is off here”? If that person is not clearly identified, or no longer has enough practice to exercise that judgement, that is where the real risk begins.
What These Questions Reveal
AI is neither a threat nor a solution. It is a revealer. It reveals the quality of internal governance, the clarity of roles, the robustness of organisational culture. The companies that fare best are not necessarily those with the most sophisticated tools. They are the ones that have consciously decided how they want to use them.
Source: SwissNova — corporate training provider in Geneva & Vaud
