Abstract

Trust in AI agents and automation is increasing, but it remains conditional and tightly bounded by context. People and organizations broadly accept AI for low‑risk, supervised tasks, yet remain skeptical of high‑stakes autonomous decision‑making. Adoption is accelerating faster than trust, creating a persistent ‘trust gap” shaped by limited governance capacity, uneven AI literacy, and concerns about safety, fairness, and accountability. Trust is strongest in predictable, controlled environments—routine automation, decision‑support systems, and bounded workflows where humans retain oversight. It declines sharply in healthcare, finance, policing, and legal contexts, especially when systems are opaque or autonomous. Public concern continues to rise around misinformation, privacy, cybersecurity, and job displacement. Public attitudes show cautious optimism. Individuals value efficiency gains but worry about transparency, oversight, and responsible use. Organizations mirror these dynamics. Leaders trust AI’s strategic importance but hesitate to grant agents autonomy. Security, risk, and unclear liability remain dominant barriers. To build trust, organizations increasingly adopt staged autonomy: AI recommends actions, humans approve them, low‑risk tasks are automated, and autonomy expands only after sustained positive performance. Hybrid human‑AI systems are becoming the dominant model, emphasizing accountability, user control, and continuous monitoring.

1) Introduction

Public trust in AI agents and automation has entered a phase defined by caution rather than confidence. Across 2025–2026, people increasingly recognize the efficiency and convenience these systems provide, yet their trust remains situational and highly dependent on context. Individuals tend to support AI when tasks are routine, risks are low, and systems behave predictably, but concerns about safety, oversight, and transparency continue to shape attitudes. As adoption accelerates globally, trust has not kept pace: many feel pressure to use AI to remain competitive, even while uncertainty about responsible use persists.

2) Public Trust in AI Agents and Automation

Public trust in AI agents and automation in 2025–2026 is best described as conditional, cautious, and highly context‑dependent. People recognize substantial efficiency gains, yet trust varies according to task type, perceived risk, system transparency, and prior experience. Surveys across multiple regions show a pattern of cautious optimism: individuals appreciate the potential of AI to streamline work and improve access to information, but concerns about safety, oversight, and transparency remain persistent. Trust is therefore not a stable attribute but a situational response shaped by how clearly an AI system communicates its purpose, how reliably it performs, and how much control users retain over its actions.

3) Global Trust Trends and the Adoption–Confidence Gap

Across 47 countries and tens of thousands of respondents, global research reveals a widening gap between rapid AI adoption and slow trust formation. Many people feel compelled to adopt AI to avoid falling behind economically or professionally, yet they remain uncertain about how to use these systems responsibly. AI literacy lags behind deployment, creating structural mistrust and reinforcing the need for human oversight.

Industry surveys deepen this picture. Responsible‑AI maturity has risen modestly, but only a minority of organizations have governance frameworks capable of supporting autonomous agents. Concerns have shifted from AI producing incorrect information to AI taking unintended actions, misusing tools, or operating beyond guardrails. A majority of organizations experiment with agents, yet security and risk concerns remain the dominant barrier to scaling. Parallel surveys show that trust increasingly depends not on the technology itself but on an organization’s readiness, control mechanisms, and governance capacity.

Socioeconomic divides further shape trust. Lower‑income and middle‑income groups often fear being left behind by generative AI, and many cite AI as a major factor influencing institutional trust. Notably, distrust is frequently anticipatory rather than experience‑based: only a small minority report negative personal encounters with AI. Hands‑on experience consistently increases trust, raising adoption enthusiasm and strengthening confidence among employees who have benefited from AI tools.

Sector‑specific findings reinforce this complexity. In healthcare, many respondents believe AI can perform certain tasks as well as or better than doctors, yet overall confidence in health decision‑making continues to decline. In the public sector, skepticism toward government use of AI persists even when deployments align with areas where citizens expect benefits. Business leaders rarely trust AI agents to run core processes autonomously, and consumers remain comfortable only with narrow, well‑understood applications such as navigation or recommendation systems.

4) Contexts of High Trust: Predictability, Control, and Clear Value

Trust is highest in controlled environments where AI’s value proposition is clear and risks are minimal. Routine automation—such as scheduling, data ingestion, and repetitive administrative tasks—benefits from deterministic behavior and well‑defined parameters, enabling organizations to rely on agents for substantial time savings. Decision‑support systems also enjoy strong trust: when AI acts as a co‑pilot rather than an autonomous decision‑maker, users appreciate its ability to retrieve information, analyze trends, and propose actions while leaving final judgment to humans.

Strategic trust is also notable. Many executives view AI as an economic necessity and actively pursue agentic AI to maintain competitiveness. Across these contexts, trust grows when systems are predictable, accurate, and transparent, and when users can verify outputs, override decisions, and observe consistent behavior over time. Strong governance, high‑quality data, and robust security controls further reinforce confidence, especially when AI demonstrably improves productivity or customer experience.

A cross‑domain comparison shows trust highest in administrative tasks and customer support, moderate in software development and financial analysis, and lowest in healthcare decision‑making, legal advice, and autonomous high‑stakes actions.

5) Determinants of Trust: Transparency, Competence, Fairness, Oversight

Research identifies several core determinants of trust in AI systems. Transparency and explainability are central: people trust AI more when they understand how it works and why it produces certain outputs. Perceived competence—reliability, accuracy, and consistency—also plays a major role. Fairness and ethical safeguards matter deeply, as bias, opaque decision‑making, and unclear accountability undermine confidence.

Cultural context shapes trust across countries and economic groups, while human oversight remains a universal preference. Trust increases when systems clearly communicate their limitations, allow human intervention, protect privacy, and operate within established governance frameworks. Public concern continues to rise around safety, societal impact, misinformation, and radicalization risks.

Automation is generally more trusted than autonomous agents, but trust remains conditional. Employees appreciate efficiency gains yet worry about job displacement, surveillance, and unclear accountability. Students use AI widely but question its fairness and accuracy.

6) Contexts Where Trust Declines

Trust drops sharply in high‑stakes contexts such as healthcare, policing, finance, and legal decision‑making. Opaque or black‑box systems exacerbate skepticism, as do AI‑generated misinformation and autonomous actions without human oversight. Cybersecurity, privacy, and misuse of sensitive data are major concerns, frequently cited as barriers to wider deployment.

Users fear biased or low‑quality outputs, hallucinations, and unclear accountability when autonomous systems cause harm. Research on human–automation trust shows that both over‑trust and under‑trust are common, especially when users misunderstand system limits. This mismatch leads organizations to restrict AI from mission‑critical workflows or customer‑facing decisions. Trust is undermined by lack of explainability, inaccurate outputs, security vulnerabilities, unfair outcomes, ambiguous responsibility, and fears of job displacement or loss of human control.

7) Barriers to Trust: Black‑Box Risks, Governance Gaps, and Consumer Skepticism

Despite rising adoption, operational and consumer trust is fracturing. The most significant fear is loss of predictability. Autonomous agents struggle with edge cases, sometimes hallucinating, bypassing guardrails, or executing compounding errors before humans intervene. This “exception problem” highlights the brittleness of agentic systems compared to human adaptability.

Oversight and accountability gaps deepen mistrust. Security and risk concerns dominate organizational hesitations, and only a minority have mature responsible‑AI controls. Liability for autonomous errors remains ambiguous. Internal trust is also strained by widespread “Shadow AI,” where employees use unsanctioned tools due to slow official pipelines, undermining security and organizational cohesion.

Consumer skepticism is pronounced. Many do not trust AI as a reliable information source, and a large majority distrust companies’ handling of personal data. Deepfakes and synthetic media erode trust in digital content, creating a baseline of skepticism toward what people see or hear online. Organizations trust agents only when boundaries are clear, human approval is required for high‑impact actions, logs are auditable, permissions are tightly controlled, and performance is continuously monitored.

8) Staged Autonomy and Hybrid Human–AI Systems

To manage risk and build trust, organizations increasingly adopt a staged approach to autonomy. AI first recommends actions, humans review and approve them, and AI automates low‑risk tasks. Only after sustained positive performance do agents receive greater autonomy.

Scholars and policymakers emphasize that trustworthy AI requires fairness, transparency, safety, and respect for human rights. International frameworks call for explainability, auditability, and mechanisms to contest or correct AI decisions. Design guidance focuses on embedding accountability, meaningful user control, and continuous monitoring. In practice, this results in hybrid systems where agents handle routine work while humans oversee high‑impact decisions.

9) Consistent Behavioral Patterns Across Surveys

Across surveys, several patterns recur. Employees frequently use AI tools but verify outputs before relying on them. Business leaders are optimistic about productivity gains but emphasize governance and risk management. Consumers trust AI more for assisting with decisions than for making them independently. Trust increases significantly after positive firsthand experience with well‑performing systems, reinforcing the importance of practical exposure.

10) Conclusion

Overall, trust in AI agents and automation is rising but remains conditional, cautious, and bounded. People and organizations trust AI for constrained, supervised tasks but withhold full trust until systems demonstrate security, fairness, transparency, and alignment with human oversight. Trust grows where users have direct positive experience and declines where autonomy increases or stakes are high.

The primary constraint on scaling agentic AI is governance capacity rather than technological capability. For AI agents to evolve from experimental tools to trusted infrastructure, organizations must implement strict identity management, permission‑based systems, and verifiable audit trails that maintain human involvement in high‑stakes decisions. Trust must be earned through demonstrated reliability, transparency, and accountability, reinforcing a model where AI augments rather than replaces human judgment.

(Researched, edited and patchworked with the aid of the AI systems ChatGPT, Google Gemini, Microsoft Copilot, and Claude. Graphs generated by ChatGPT, illustration by Google Gemini.)

Thorsten Koch, MA, PgDip
Policyinstitute.net
12 July 2026