Employee distrust is slowing AI at scale

Employee distrust could become the real constraint on autonomous AI

AI adoption is moving forward, but there is a clear trust gap inside companies. Managers and employees are becoming more comfortable using AI, yet the people leading adoption tend to be more positive about the technology than frontline workers. They are also more willing to let AI take over parts of the workflow.

That difference matters as companies move from AI that recommends to AI that acts. Traditional software generally follows instructions defined by people. Autonomous AI can go further: it can make decisions and take actions without waiting for approval at every step. For employees, that creates a basic question: How much control am I giving up, and what happens if the system gets something wrong?

Only 11% of employees said they were comfortable with AI taking actions they had not reviewed and approved. Employees reported greater personal risk from autonomous systems because they have less control over how and when those systems are deployed.

Trust can become an operational constraint. A company can deploy sophisticated AI, but if employees repeatedly verify its work, avoid using autonomous features, or escalate routine decisions back to humans, the expected productivity gains will be smaller.

The solution is not to demand more trust. Companies need to create conditions that justify it. Employees need to understand what an AI system is authorized to do, what it cannot do, how decisions can be reversed, and who is accountable when something fails. Leadership also needs feedback from the people who actually use these systems. They often encounter weaknesses that do not appear during controlled testing.

The opportunity remains significant. Autonomous AI can remove repetitive work and accelerate decisions. But technical capability alone will not determine how quickly companies capture that value. The organizations that make autonomy understandable, controlled, and accountable will have a much stronger foundation for scaling it.

Employees want AI assistance, but they still want control over important outcomes

Workers are already using AI. Their behavior suggests that the main issue is not resistance to the technology itself. It is the level of authority employees are willing to give it.

AI is most commonly used once or twice a day. Employees generally use it to understand information, generate a starting point, or complete routine tasks. These activities share an important feature: a person can inspect the result before using it.

This preference gives executives useful information about how AI should be deployed today. Employees appear more comfortable when AI accelerates work without removing human judgment. A generated summary can be checked. A draft can be edited. A routine output can be validated. Fully autonomous action changes the risk because an incorrect result can become an incorrect action before anyone intervenes.

Human oversight should also be based on risk rather than applied to every task. Requiring approval for every low-impact AI action can eliminate much of the efficiency automation is supposed to create. At the same time, giving AI unrestricted authority over financially, legally, operationally, or reputationally significant decisions can introduce unnecessary exposure.

Executives can address this by defining levels of autonomy. Low-risk, reversible actions can receive pre-approval. More consequential actions can require human confirmation. As systems demonstrate reliable performance, organizations can expand their authority while continuing to monitor outcomes.

This is a practical route to scale. Start where AI creates measurable value and the consequences of failure are manageable. Establish evidence that the system performs reliably. Then increase autonomy as confidence, governance, and technical performance improve together.

Enterprises are moving toward autonomous AI faster than trust is developing

Companies are preparing to give AI much more authority. The strategic direction is clear: AI is moving beyond generating content and recommendations toward making decisions and taking actions with limited human involvement.

The problem is that confidence has not advanced at the same speed. According to a June survey published by Kyndryl, only 25% of business and technology leaders said they completely trust their AI systems. Yet more than four-fifths said they expect autonomous AI agents to make decisions with material business impact within the next year.

That gap deserves executive attention. Companies are preparing to give AI greater control over meaningful business activities even though most leaders do not yet express complete trust in the systems involved. Once autonomous AI affects customers, finances, operations, security, or regulated processes, errors can create consequences beyond a single incorrect output.

This does not mean companies should wait for perfect AI. That standard is unrealistic. Human decision-making is not perfect either. The more useful objective is to define acceptable performance and risk for each use case. Leaders need evidence that a system performs reliably within its assigned scope and mechanisms to identify, contain, and reverse failures.

The pace of deployment should therefore depend on the consequences of an error. An AI agent handling a low-risk internal process can operate with greater freedom than one approving financial transactions, modifying critical systems, or communicating consequential information to customers. Autonomy should be a business decision based on measurable risk, not simply a feature that is switched on because the technology supports it.

There is also an economic issue. Increasing autonomy can reduce repetitive work, accelerate processes, and allow employees to concentrate on higher-value decisions. But weak controls can create new costs through manual corrections, security incidents, compliance failures, and damaged customer confidence. Executives need to measure both sides of that equation.

The companies that manage this well will not necessarily be those deploying autonomous AI everywhere first. They will be the organizations that identify where autonomy produces meaningful value, establish clear performance requirements, and expand deployment when the evidence supports it.

Weak governance and unclear accountability are undermining trust in autonomous AI

Trust in AI is closely connected to governance. Employees need to know what a system is permitted to do, what information it can access, who monitors its actions, and who is responsible when something goes wrong. Without clear answers, autonomous AI creates uncertainty for both employees and management.

Employees worry that AI could expose sensitive information, fail to understand the context needed to produce an appropriate result, or take actions beyond its intended scope. About one-fifth also said they were concerned that they could be held responsible for mistakes made by AI.

For executives, accountability is particularly important. An employee should not have to determine after an incident whether they, their manager, the technology team, a vendor, or another business function owns the failure. Responsibility needs to be established before an autonomous system is deployed.

Governance also needs to extend beyond policy documents. Companies should define which data an AI system can use, the actions it is authorized to perform, the conditions that require human approval, and the circumstances under which autonomous operation must stop. They should also maintain records that make important AI actions traceable. For high-impact use cases, the ability to reverse or correct an action should be designed into the process from the beginning.

This matters because governance is not simply about reducing risk. Done well, it can enable faster adoption. When boundaries and responsibilities are explicit, teams can give AI greater authority in approved areas without repeatedly debating who can make each decision.

The executive objective is therefore to make autonomy controlled and measurable. Define authority before deployment, assign accountability clearly, monitor what the system actually does, and create processes for intervention when performance moves outside acceptable limits. With those foundations in place, organizations can increase AI autonomy while keeping people in control of the decisions that carry the greatest business consequences.

Autonomous AI should expand through clearly defined levels of authority

Giving an AI system autonomy should not be an all-or-nothing decision. Organizations can define exactly which actions a system may complete independently and which require human approval. Start with low-risk, pre-approved actions, while retaining human review for decisions with greater consequences.

This creates a practical path to adoption. Routine, reversible tasks can receive more autonomy when the potential cost of an error is low. Actions involving sensitive information, significant financial commitments, regulatory obligations, security, employees, or customers may require stronger controls. The level of human involvement should reflect the business impact of a possible failure.

Employees should have a role in defining these boundaries. They understand workflows at an operational level and can identify risks that may not be obvious during system design. Their participation also makes it clearer what authority the AI has, when an employee is expected to intervene, and where responsibility lies when a system makes a mistake.

Leaders should make these rules specific. Employees need to know what the AI can access, which actions it can execute, when approval is mandatory, and how an autonomous action can be stopped or reversed. Staff need confidence that autonomous systems are secure, explainable, reversible, and governed by people.

The executive challenge is to preserve the economic benefits of automation without adding unnecessary human approvals. Reviewing every AI action reduces the value of autonomy. Removing oversight too aggressively increases operational and governance risk. Companies need to determine where human judgment creates real value and where pre-approved automation can safely take over.

That structure can evolve. As an AI system demonstrates reliable performance, the organization can authorize additional actions or reduce approval requirements. If performance deteriorates or operating conditions change, those permissions can be restricted. Autonomy becomes something the business actively manages rather than a permanent technical setting.

Scale autonomy after proving value, while distributing risk fairly

The strongest case for wider AI autonomy comes from demonstrated performance. Instead of deploying autonomous systems broadly from the beginning, organizations can establish specific use cases, define acceptable outcomes, measure results, and expand authority when the evidence supports doing so.

Mei Dent, Chief Product and Technology Officer at TeamViewer, captured this approach directly: “Autonomy will scale when organizations can prove it works in specific, trusted use cases, then expand it with confidence.” Dent also noted that AI systems and agents change the traditional software model because they can do more than simply execute explicit instructions from users.

For executives, proving that a use case “works” should mean more than showing that the AI can complete the task. Organizations should assess accuracy, reliability, security, business impact, exception rates, and the frequency of human intervention. For consequential applications, leaders should also consider whether actions are traceable and reversible and whether the system remains within its authorized scope.

Risk distribution is equally important. The consequences of autonomous AI should not fall disproportionately on one employee, team, or workplace group. This is particularly relevant when employees may be held accountable for decisions they did not fully control. About one-fifth of employees expressed concern that they could be held responsible for AI mistakes.

Clear ownership can reduce that problem. Business leaders should determine who owns the AI-enabled process, who monitors system performance, who handles exceptions, and who carries accountability when the technology operates as authorized but still produces an undesirable outcome. Technology teams, business units, risk functions, and individual employees may each have responsibilities, but those responsibilities need to be explicit.

Scaling should then follow evidence. A company can begin with narrowly defined activities where outcomes are measurable and risks are manageable. Strong performance can justify broader permissions, more complex tasks, or deployment across additional parts of the organization. Poor performance should trigger changes to the system, tighter controls, or reduced autonomy.

The objective is not maximum autonomy. It is useful autonomy. The broader opportunity is to use AI and automation to “prevent disruption, improve digital experiences and free people to focus on higher-value work.” For C-suite leaders, that means measuring success by business outcomes and controlled risk, rather than by how many processes can operate without human involvement.

Key executive takeaways

• Trust is the constraint on AI scale: Only 11% of employees are comfortable with AI acting without their review or approval. Leaders need to build trust alongside technical capability if they want autonomy to deliver real operational value.
• Preserve human control where it matters: Employees favor AI for tasks they can review, such as understanding information, creating drafts, and handling routine work. Use risk-based human oversight while allowing safe, reversible tasks greater autonomy.
• AI autonomy is moving faster than confidence: A Kyndryl survey found that 25% of business and technology leaders completely trust their AI systems, yet more than 80% expect autonomous agents to make materially significant decisions within a year. Scale authority according to demonstrated reliability and business risk.
• Governance enables greater autonomy: Concerns about sensitive data, contextual errors, unauthorized actions, and personal accountability can undermine adoption. Define AI permissions, accountability, monitoring, and intervention processes before expanding deployment.
• Give AI clearly defined authority: Pre-approve low-risk autonomous actions while requiring human review for higher-impact decisions. Involve employees in setting these boundaries and expand permissions as systems establish reliable performance.
• Prove value before scaling autonomy: Start with measurable, trusted use cases and expand only when performance supports it. Distribute accountability fairly across teams and measure success by business outcomes, reliability, and controlled risk rather than maximum automation.