There is a simple pattern running through nearly every major workplace AI survey published this year: the higher up you are in an organization, the more you trust AI. Executives love it. Frontline workers are much more cautious. And the companies that ignore this gap are spending money on AI that their own people are quietly working around.
The numbers are stark. A Dayforce survey of nearly 7,000 workers across six countries found that 87% of executives use AI at work, compared to just 27% of workers. The same study found that 71% of workers had received no AI training in the past year. You cannot ask someone to trust a tool they have never been taught to use.
The trust problem gets sharper when AI moves from giving suggestions to taking action. Deloitte's research found that employee trust in AI systems that act independently, without asking first, dropped 89% over just two months in mid-2025. That is not a slow erosion of confidence. That is a collapse. And it is happening at the exact moment when companies are doubling down on giving AI more autonomy.
The Kyndryl People Readiness Report, published in June 2026 and based on 1,100 senior leaders across eight countries, found that 81% of organizations expect AI agents to be making decisions with real business consequences within the next year. Only 25% said they fully trust those systems to operate without a human checking the work. Companies are planning for a future that their own people do not yet believe in.
Why the gap? Workers are not being irrational. When AI takes an action automatically and it turns out to be wrong, someone has to answer for it. In most organizations, that someone is the person closest to the customer or the process, not the manager three levels up who approved the deployment. Research cited in the source survey found that about one in five employees worry they will be personally held responsible for AI mistakes. That is a legitimate concern, and it shapes behavior.
BCG's annual AI at Work survey, which covers close to 12,000 employees across more than a dozen countries, found that frontline workers have hit a ceiling in AI adoption. The share of frontline employees who regularly use AI has stalled at around 51% while manager usage has climbed to 78%. When workers do use AI, they tend to use it for tasks where they can review the output before it matters, drafting, summarizing, finding information. The things where a mistake stays private.
The fix is not more AI. It is more honesty about how AI is being deployed and who carries the risk when it fails. Organizations that have made real progress on trust share a few common practices. They define clearly which tasks AI can handle without asking anyone. They define equally clearly which tasks need a human sign-off. And they involve employees in drawing those lines, rather than announcing them from above.
McKinsey has noted that nearly 80% of companies are using some form of AI, yet more than 60% report no meaningful impact on their bottom line. The most common reason: not enough employees have the skills or confidence to work effectively alongside the technology. Training matters, but so does giving people actual control over what the AI does in their area of work.
The companies that will get real value from AI agents are not the ones deploying the most of them. They are the ones that start with a handful of low-risk, well-defined tasks, prove the system works, and then expand from there with their employees watching, involved, and willing. Trust is built in small steps, and it is lost quickly when something goes wrong with no explanation and no accountability.