The Trust Gap: Why Fewer People Believe in Full AI Autonomy

Explore the psychology behind falling trust scores, even as more systems act on our behalf.

Welcome Back,

If you only listened to product launches, you might think we are racing toward a world where AI handles everything for us, from driving cars to running companies. In practice, people seem much less eager to hand over the keys completely.

Recent global surveys show adoption in AI is growing fast, but trust is not keeping pace. An EY sentiment report found that 84 percent of respondents used AI in the past six months, yet two thirds still say human oversight remains essential and worry about hacked or breached systems, while nearly three quarters fear they will not be able to tell what is real or AI generated, according to EY's 2026 AI Sentiment Report.

A separate global study on trust and attitudes in AI found that only about 46 percent of people worldwide are willing to trust AI systems, even though 66 percent use AI regularly, according to Trust, Attitudes and Use of AI: A Global Study 2025.

Today we dive into what that trust gap really is, why full autonomy triggers discomfort even when basic AI tools feel helpful, and what psychology research can tell us about how trust might be rebuilt.

📉 Adoption Is Rising, Trust Is Not

The headline trend is surprisingly simple. People are using AI more than ever, and they are letting it act more autonomously than they might freely admit. Yet when asked whether they trust that autonomy, they hesitate.

The EY 2026 AI Sentiment Report found that 16 percent of respondents globally now use AI systems that act without human intervention, meaning fully autonomous systems have already moved beyond theory into everyday reality, even if most people still describe their use as experimental or low stakes. At the same time, 66 percent worry about AI systems being hacked or breached and almost seven in 10 say human oversight remains essential, according to EY's survey.

A global trust study led by researchers in Australia previously found that three out of five people are ambivalent or unwilling to trust AI, and three out of four are concerned about the risks associated with AI, with cyber security and loss of privacy at the top of the list, according to Trust in Artificial Intelligence: A Global Study.

That pattern matters because it suggests the trust gap is not about capabilities. It is about control, accountability, and safety, the human side of autonomy rather than the technical side.

Surveys show a global tension: people use AI frequently, but many remain ambivalent or uneasy about trusting it fully.

🧠 The Psychology Behind Distrust in Autonomy

Psychology research on trust in AI suggests that people rarely think about AI in isolation. They think about who is using it, who is responsible, and how that affects their own autonomy. A recent review in the journal Public Trust in Artificial Intelligence Users argues that people form trust judgments about AI users the same way they do about other humans, based on perceived ability, benevolence, and integrity, not just technical performance, according to Dang et al., 2024.

At the same time, ethics researchers have pointed out that AI mediation of human activities can influence how we make decisions and behave, sometimes undermining our sense of autonomous deliberation. A literature review on AI and human autonomy notes that people worry when AI systems seem to override or nudge their choices without clear consent, which can feel like a threat to personal agency, according to AI and Ethics, 2026.

A separate analysis from the Hertie School argues that many perceived threats to autonomy are not inherent to AI itself, but to design choices: lack of transparency, poor consent, weak privacy controls, and unclear accountability can make people feel that AI systems disrespect or erode their autonomy, according to The threat to human autonomy in AI systems is a design problem we can fix.

In other words, many people are less afraid of intelligent systems than they are of unaccountable ones. Full autonomy feels risky when it seems to remove the last human checkpoint from the loop.

⚖️ Trust Depends on the Context, Not Just the Tech

One of the most important findings across global surveys is that trust in AI is highly context dependent. People are far more comfortable with autonomous AI in some domains than others, and that shapes how they answer broad questions about autonomy.

The Trust in Artificial Intelligence global study notes, for example, that AI use in healthcare is generally more trusted than AI use in hiring, and that people tend to have more faith in AI's capability and helpfulness than in its safety, security, and fairness, according to UQ's global study.

A 2025 analysis of public trust in AI cognitive capabilities found similar patterns, with respondents more willing to trust AI to perform narrow, well defined tasks than to handle broad, open ended decisions. Trust scores were higher when the AI's role was clear and bounded, according to Nature Scientific Reports, 2025.

Put differently, people are more willing to accept autonomy when its scope is narrow, the stakes are clear, and a human remains visibly in charge of the broader decision.

People tend to trust autonomy more when AI is clearly supporting a professional, not replacing them.

🏛 Design and Governance: Where Trust Is Won or Lost

Multiple studies emphasize that trust is not just about the AI system itself, but about the institutions and safeguards around it. The global trust study outlines four pathways that strengthen public trust in AI: institutional safeguards and regulation, perceived benefits, risk reduction, and user knowledge. Of these, institutional safeguards are the strongest driver of trust, according to UQ's research.

A blog from the Hertie School goes further, arguing that there is no intrinsic conflict between human autonomy and AI. The tension arises when systems are designed without transparency, consent, or respect for what researchers call meta autonomy, our ability to decide when to delegate a decision and when to retain control. Fixing those design choices can restore autonomy without abandoning AI altogether, according to Hertie School's analysis.

McKinsey's 2026 State of AI Trust report echoes this, noting that responsible AI practices and clear governance have become foundational requirements for realizing AI's potential at scale, not optional extras, according to McKinsey's analysis.

🚦 Should AI Ever Be Fully Autonomous?

Not all autonomy is the same. A recent position paper called AI Must Not Be Fully Autonomous argues that systems capable of developing their own objectives without responsible human oversight cross an important boundary. The authors suggest that fully autonomous AI of that kind should be avoided, and that responsible human supervision must remain in the loop for high level goals and accountability, according to Alruwaili et al., 2025.

Taken together, surveys and ethics research point toward a nuanced answer. People are willing to accept significant autonomy in narrow, well regulated domains, especially when it clearly serves their interests and remains easy to override. They are far less comfortable with open ended autonomy that feels unaccountable, opaque, or misaligned with human values.

In that sense, the trust gap is both a warning and a design brief. It tells us people are not ready for full autonomy across the board, but it also hints at where autonomy might fit: as a careful, transparent partner rather than a silent replacement.

🔗 Sources and Further Reading

Thanks for reading AI Spotlight.

Our mission is simple: deliver clear, trustworthy, and actionable AI insights that help professionals stay ahead without the hype.