What Happened
This week, a figure was released that should give us pause: 51 percent of Americans use AI for research. Only 21 percent trust AI-generated information. The Quinnipiac survey from April 2026 thus confirms what various studies have been suggesting for months: We have collectively adopted a tool that the majority of us distrust—and yet still use every day.
This is not an American phenomenon. According to ManpowerGroup, 41 percent of employees in Germany use AI regularly in their daily work. At the same time, these same studies describe an “AI Confidence Gap”—the ability to critically evaluate one’s own AI outputs is growing more slowly than the use of AI itself. And news from Switzerland reads like a all-clear: the predicted AI job apocalypse has not materialized so far. But lead time is not the same as immunity.
Three data points, one pattern: AI is used because it is useful. Not because people trust it.
What the world is saying about it
The mainstream discourse currently revolves around two poles. On one side are the techno-optimists: using AI without full trust is rational—after all, we use elevators without understanding the physics behind them. AI is a tool like any other, and skepticism is a healthy sign of maturity. On the other side are the cautionary voices: those who trust systems they don’t understand make themselves an extension of an uncontrolled technology.
Both are right. And both miss the point.
The more interesting observation comes from research: Martinetti Maggetti of the University of Lausanne describes a transparency dilemma in his paper “Reciprocal Trust and Distrust in AI Systems.” More technical information about AI reduces trust rather than strengthening it. The more one knows about training data, confidence scores, and hallucination rates, the more uncertain one feels. This is why the typical tech industry response—more explainability, more transparency, more documentation—does not structurally solve the trust problem.
As for the governance debate: A study by Old Dominion University and Deloitte finds that 76 percent of surveyed data leaders cannot monitor their employees’ use of AI. AI runs in production systems without compliance teams’ knowledge. “Shadow AI” is not just a security risk—it is a sign that the speed of adoption has overwhelmed governance capacity.
What we think about this
The trust-usage paradox is not an anomaly. It is the result of a rational cost-benefit analysis under uncertainty. Anyone who uses ChatGPT for a first draft and double-checks the result is acting reasonably—even if they don’t “trust” AI. The real danger lies not in mistrust, but in the two extremes: automation bias (trusting too much) and algorithm aversion (discarding everything after a single error).
What really concerns me about the current data is the governance gap. A 76 percent oversight gap among data leaders is not a figure we can simply brush aside. Shadow AI in production systems means: decisions are made based on AI outputs for which no one takes responsibility. That is not the problem with AI—it is the problem of management, which has driven AI adoption without providing structural answers to the question: Who checks what, under what conditions?
For companies in the DACH region, there is a concrete message here: The “AI Confidence Gap”—where usage grows faster than critical expertise—is not an inherent characteristic of the technology. It is the predictable consequence of rollouts without training in critical evaluation. Anyone who implements AI tools without simultaneously investing in the ability to question AI outputs creates automation bias at the corporate level. That is more expensive than the tool license.
The concept of “Watchful Trust” is the only sensible framework for dealing with this situation. Treat AI outputs as drafts, not as final results. Build systemic trust through repeatable positive experiences, not through marketing promises. And define clear organizational boundaries: In which contexts do we trust AI outputs directly, in which do we verify them, and in which do we not use them at all?
These are questions that no AI can answer. They are leadership tasks.
The Verdict
Conclusion in one sentence: Using AI without trust isn’t the problem—a lack of governance and the gap between adoption speed and critical expertise certainly are.
The AI Report is published every Friday. Subscribe for free:
👉 aisyndicate.ch/#/portal
