AI Adoption and Trust: Why Confidence is Struggling to Keep Up
Artificial intelligence is finding its way into workplaces and everyday life at remarkable speed. But beneath the excitement, a growing body of research points to a problem that technology alone cannot solve: people are still not convinced they can trust it.
Every time we browse the news these days it seems there is no shortage of enthusiasm for artificial intelligence. Businesses are investing, and governments are developing AI strategies, and new tools which promise to transform everything from healthcare to education, and journalism. Yet behind the headlines, the reality is more complicated.
AI may be becoming increasingly available, but using it confidently, responsibly and at scale is another matter altogether. The question is no longer simply whether AI can do something. It is whether people trust it enough to let it. And increasingly, the evidence suggests that trust is struggling to keep pace with technological progress.
The numbers tell an interesting story
According to Eurostat figures published in December 2025, just 20% of EU enterprises with at least 10 employees were using AI technologies in 2025. This represents significant growth from 13.5% the previous year, but it also means that four out of five businesses had yet to report using AI. Adoption also varies dramatically across Europe. In Denmark, 42% of enterprises reported using AI, compared with just 5.2% in Romania.
These differences cannot simply be attributed to trust. Investment, digital skills, organisational readiness and access to technology all play important roles. But trust is increasingly emerging as a critical part of the equation. A major 2025 global study by the University of Melbourne and KPMG, involving more than 48,000 people across 47 countries, revealed a striking contradiction. While 66% of respondents said they were intentionally using AI with some regularity, only 46% were willing to trust it. In other words, people are using a technology that many of them remain deeply uncertain about.
Professor Nicole Gillespie, who led the research, highlighted the importance of public confidence:
"The public’s trust of AI technologies and their safe and secure use is central to sustained acceptance and adoption."
This is a very important distinction. Using AI does not necessarily mean trusting it. Someone might use a chatbot to engage with a service, or use AI to draft an email but hesitate to rely on Aitwhen making a medical, financial or professional decision. And that hesitation may be entirely justified.
Why are people still reluctant to trust AI?
There is a tendency among governments to assume that resistance to AI comes from a lack of understanding. If people knew more about how the technology worked, perhaps they would be more willing to embrace it. But in the THEMIS 5.0 project we believe the evidence points to something more complex.
The Stanford AI Index Report 2026 found that global optimism about AI was rising. Some 59% of respondents believed AI products and services offered more benefits than drawbacks in 2025. At the same time, 52% said AI products and services made them nervous. Its clear that people can recognise AI's potential while simultaneously worrying about its consequences.
Those concerns are not abstract. They involve questions that matter in everyday life. What happens when an AI system makes a mistake? Who is responsible? Can its decisions be explained? Is personal information being protected? And can people challenge decisions they believe are wrong?
Research from Edelman's 2025 Trust Barometer AI study illustrates just how closely trust and acceptance are connected. In the UK, 43% of people who trusted AI embraced its growing use, compared with just 2% of those who distrusted it. Similar differences appeared in the United States, Germany and Brazil. The findings do not prove that trust alone determines adoption. But they underline how difficult widespread acceptance becomes when confidence is missing.
The gap between experimenting with AI and actually relying on it
The challenges become particularly visible when organisations try to move beyond small-scale experimentation. In October 2026, Reuters reported that businesses were struggling to scale AI projects, despite promising financial returns.
The report, drawing on research by consultancy BearingPoint, found that nearly three-quarters of surveyed businesses had seen positive financial results from their AI investments. Yet fewer than one-third had moved beyond pilot projects.
Regulatory challenges and difficulties integrating AI into existing IT systems were among the obstacles identified.
This is an important reminder that adoption is not just a question of whether AI works. Organisations also need the structures, skills and safeguards to use it effectively.
A successful pilot does not automatically translate into a system that employees understand, managers can oversee or customers feel comfortable relying on.
And when AI starts influencing decisions with real consequences, those questions become much harder to ignore.
Trust means different things in different situations
Consider a doctor using AI to help identify abnormalities in a medical scan. Accuracy is essential, but so is understanding why the system has reached a particular conclusion and knowing when its recommendation might be unreliable.
Now consider a port operator using AI to improve logistics and predict operational disruptions. Here, reliability, robustness and the ability to respond to unexpected situations may be the overriding priorities.
For a journalist using AI to support research or content production, the concerns may be entirely different. Can sources be verified? Could the system introduce bias? How can the journalist remain accountable for information that AI has helped produce?
These are all questions about trust, but they are not asking for the same thing. The media sector provides a particularly revealing example. The Reuters Institute's Digital News Report 2025 found considerable public unease about AI-generated news. Respondents were more likely to believe generative AI would make news less trustworthy, less accurate and less transparent than the reverse. And as the 2026 Digital News Report shows, overall trust in news has fallen to 37% across the markets surveyed, its lowest recorded level since the Institute began tracking it in 2015.
In an environment where public confidence is already fragile, the introduction of AI needs particularly careful consideration. What makes AI trustworthy in one setting may not be enough in another.
AI adoption and trust: Moving from asking for trust to demonstrating trustworthiness
This is where the conversation around AI needs to change. For too long, discussions about trustworthy AI have centred on general principles. Fairness, transparency, accountability, safety and explainability are all essential, but what do they actually mean for someone using an AI system in a real working environment? How can an organisation establish whether a particular system is trustworthy for a particular task? And how can it identify and address weaknesses before those weaknesses affect people?
These are among the questions being explored by our EU-funded THEMIS 5.0 project.
Rather than treating trustworthiness as a fixed characteristic that an AI system either possesses or lacks, THEMIS 5.0 takes a human-centred, context-sensitive approach.
It recognises that different users have different priorities, and that assessing trustworthiness requires looking beyond technical performance to consider human values, decision-making, fairness, risk and the potential consequences of AI-generated recommendations.
Through our work across healthcare, ports and journalism, we have developed approaches and tools to help organisations assess and improve AI trustworthiness in practical settings. Our ambition was not simply to persuade people to trust AI more. It was to help ensure that AI systems deserve the level of trust placed in them. Because encouraging trust without demonstrating trustworthiness risks creating a different problem, people relying too heavily on systems whose limitations they do not fully understand.
Trust cannot be an afterthought
AI adoption will almost certainly continue to grow. The potential benefits are too significant for organisations and individuals to ignore. But as we have seen in this article technological progress does not automatically produce public confidence. The evidence emerging from recent research points to a fundamental challenge. People are increasingly willing to experiment with AI, yet remain cautious about the technology's reliability, accountability and wider consequences.
That caution should not simply be dismissed as resistance to innovation. It can also be an opportunity to ask better questions, design better systems and make more informed decisions about where and how AI should be used. The next stage of AI adoption will require more than powerful models, ambitious strategies and successful demonstrations. It will require organisations to show that their AI systems are suitable for the people, decisions and environments in which they operate.
The future of AI will not be determined by capability alone. It will also depend on whether we can make trustworthiness something that can be assessed, demonstrated and continuously improved.
That is the challenge at the heart of THEMIS 5.0. Check out our papers and deliverables to learn more.
Comments