Summary
- In the multinational survey, 46% of middle managers wanted human involvement versus 24% of executives.
- Security and privacy concerns were cited as an AI adoption barrier by 45% of IT decision-makers.
- The study surveyed 4,200 respondents across nine countries, including four European markets.
Middle managers are more likely than senior executives to want human involvement when artificial intelligence makes decisions affecting IT security, data or business operations, according to new multinational research commissioned by TeamViewer.
The study found that 46% of middle managers wanted human oversight in those circumstances, compared with 24% of executive leaders. The figures come from a survey of 4,200 respondents conducted by Sapio Research in April and May 2026 across nine markets, including the UK, France, Germany and Italy.
The comparison highlights differences in attitudes to autonomous tools, but the published headline results should not be described as a Europe-only survey. Responses also included participants in the United States, Australia, Japan, Singapore and Brazil, with employees and managers represented across multiple industries.
Almost one third of workers cited security risks or unauthorised access among their leading concerns about AI acting without human review. Senior and middle managers each recorded 33% on that measure, compared with 23% of executives.
Access restrictions were another common consideration. TeamViewer reported that 42% of senior managers, 37% of middle managers and 32% of executives would become more comfortable with autonomous problem-solving if limits were imposed on the systems and information AI could reach.
Those differences are operationally relevant because AI tools that merely recommend a fix present a different exposure from agents authorised to change a device configuration, reset credentials or modify a business workflow. The extent of access and the reversibility of an action determine the possible consequences of error or misuse.
The survey found that 73% of workers were more inclined to trust AI when its activities and reasoning were visible, while 70% were comfortable with action provided they could intervene. Those preferences are self-reported attitudes and do not establish that any particular interface will deliver effective technical oversight.
Accountability was less settled. Only 49% said it was clear who would be responsible if an autonomous system made a mistake, with respondents dividing responsibility between IT, users and other parties. Contractual obligations and regulatory duties do not necessarily follow those individual perceptions.
Employees in energy, utilities, and oil and gas showed stronger preferences for human involvement than those in marketing and communications, at 49% and 28% respectively. These results suggest different risk tolerances across sectors, although the figures alone do not show whether the organisations involved have different technical controls.
Other responses show that concern rises as the proposed system gains more independence. Restricting what the AI can access, recording actions and providing a way to intervene address different aspects of the same governance problem. A logging mechanism may improve accountability after an action, while access restrictions reduce the resources a system can affect beforehand.
Sector differences also appear in the findings. Workers in energy, utilities and oil and gas showed greater demand for human intervention than those in marketing, media and communications. The report does not establish that one sector’s systems are inherently less secure; tasks, business consequences and workforce exposure differ.
The research found that 73% of respondents were more inclined to trust an AI system when it showed what it was doing and why. Another 70% said they were comfortable with autonomous action when intervention remained possible. Those preferences distinguish visibility and reversibility from an unconditional willingness to delegate security-sensitive work.
Energy, utilities, and oil-and-gas respondents showed a stronger preference for human involvement than respondents in marketing, media, and communications, at 49% and 28% respectively. The findings concern attitudes to AI resolving workplace IT problems, rather than an observed difference in incident rates between industries.
Sapio Research surveyed 4,200 people for TeamViewer during April and May across nine countries, including the UK, France, Germany, and Italy. The publication’s aggregate percentages should not be represented as Europe-only measurements, and beliefs about accountability do not supersede organisations’ legal or contractual responsibilities.
Accountability responses warrant particular caution: opinions about who should take responsibility are not a statement of legal duties. Organisations remain subject to applicable employment, data protection and sectoral requirements regardless of whether a survey respondent believed a machine’s mistakes should be attributed to a person or department.
