As AI moves from advice to authority, who defines its limits?

When it comes to everyday people, AI’s biggest breakthrough hasn’t been intelligence but convenience. By taking care of small, everyday tasks, AI has slipped into daily routines with little resistance. Route planning, movie suggestions and customer support are activities that people expect technology to handle. These are tasks where outcomes are easy to review, correct or override and that sense of control matters.

The 2026 survey shows that experiences like these are the building blocks of confidence. As people interact with AI in ways that feel helpful and low effort, their hesitation gives way to acceptance. AI no longer feels novel or experimental, it’s simply useful.

AI’s growth is not evenly distributed across all activities, but the direction of travel is clear. In most cases, people are not seeking innovation for its own sake. They are looking for faster service (57%), lower prices (57%) and products and services that work better (42%). Interestingly, people’s desire to make fewer decisions is not such a motivating factor (25%).

In retail, AI is increasingly being used to help people articulate intent rather than navigate filters. Instead of searching by keywords or categories, people can describe what they are trying to achieve and let AI narrow the field.

Last year, Nike launched NikeAI beta in their mobile app in the US. The updated app allows customers to describe what they want to do, such as train for a race, equip a team, or find products that meet specific preferences. AI in the app interprets that intention to make the best possible recommendation. The experience is designed to be simple, intuitive and easy to override, using large language models (LLMs) customized with Nike’s domain knowledge. By giving customers support when they need it but leaving final decisions firmly in their hands, assistive AI reduces friction, saves time and builds confidence without introducing risk.

Such situations may appear low stakes but are strategically important. Everyday interactions such as these help people learn whether a system feels reliable, understandable and aligned with their needs. They also allow organizations to gain early insights into where confidence is building, and where it can break down. For leaders, it means that they can use lower-stakes, assistive-AI experiences as a diagnostic tool. The instances reveal which tasks people are willing to hand over, what safeguards they expect and how quickly comfort can grow once value is proven.

Health-related uses stand out because they begin to move beyond convenience into more consequential territory. Many people now use AI to describe symptoms, ask what might be wrong or decide whether to seek care. Roughly 230 million people ask health-related questions of ChatGPT each week across the globe.¹

AI-powered symptom assessment apps like Buoy Health allow users to describe symptoms in their own words and tap into clinical insights. Buoy offers guidance toward what kind of care users may need. These tools are increasingly shaping real-world health decisions, including whether and when people enter formal care pathways.

Access to these kinds of tools helps explain why 26% of survey respondents report asking AI to diagnose symptoms, up from 19% last year. The same number (26%) use AI to look up health information without seeing a doctor. “What’s striking is how quickly AI is becoming a first stop for health questions,” says Kim Dalla Torre, EY Global Health Sector Leader. “As people begin to rely on it to guide real decisions, the need for safe, accurate and easily validated AI becomes far more important.”

Organizations should not rush toward full autonomy, but they should pay close attention to experiences that are already shaping expectations. It’s these moments that will determine how far, and how fast people are willing to go.

As AI becomes more embedded in everyday life, and in some cases begins to act autonomously, trust has not kept pace with capability. People remain concerned — even in Pioneer Markets — about security, control, accountability and authenticity. But interestingly, these concerns are not slowing adoption. They are shaping the way people want autonomy to show up — a critical distinction for leaders navigating the next phase of AI adoption.

Security is people’s biggest fear. Two-thirds worry about AI systems getting hacked or breached, and less than half trust governments or companies to protect personal data used by their AI systems. As autonomy increases, this concern becomes more than just a background worry. The more AI can do, the more security comes into focus. Leaders must consider what could happen if systems are compromised and how quickly that damage could scale.

People’s other major concern is control. They worry that decisions made by AI won’t reflect their personal values or priorities; seven in 10 agree that human oversight remains essential. This is not a rejection of AI but a request for agency and an ability to understand what the system is doing, intervene when it matters and opt out when stakes rise.

Six in 10 people worry about organizations failing to hold themselves accountable for AI use that leads to negative consequences. Almost as many (55%) worry that organizations will fail to comply with their own AI policies or relevant government regulations.

Some organizations are responding by making their AI commitments more explicit. For example, Tokio Marine Holdings, an insurance multinational, adopted a public AI policy as part of using AI for claims processing, fraud detection and for other customer service activities, to earn trust in the highly regulated insurance industry. Released in 2025, the policy states that the company will not rely on AI data or techniques that could introduce biases against any group, and that AI-based functions and decisions will always keep people in the loop. It further states that the company will prevent AI use that violates human rights, distributes false or biased information, or leads to leaks, falsifications or unauthorized use of personal information.

An overwhelming majority of respondents agree that AI outputs should clearly show when something is created or influenced by AI. Nearly three quarters worry that, as generative AI (GenAI) becomes more widespread, they will no longer be able to tell what is real or fake.

“The blurring of reality is one of the most underestimated AI risks,” says Katherine Boiciuc, EY Chief Technology and Innovation Officer for the Oceania region. “When people can’t easily tell what’s real, what’s generated, or who’s behind a decision, it creates a low-grade anxiety that erodes confidence — even when the technology itself is working as intended.”

Respondents also want AI tools to include parental controls and age limits and believe that countries and regions should have stronger rules governing organizational use. Across all three market types — Pioneer, Transitional and Lagging — people believe more safeguards should be enacted.

“What surprises me isn’t that people want stronger AI safeguards — it’s how consistent that expectation is across markets,” says Sarah Liang, EY Global Responsible AI Leader. “Regardless of where people live or how advanced AI adoption is, they’re asking for clearer rules, stronger accountability and visible protections.”

AI capability is advancing rapidly, while governance and accountability structures are clearly struggling to keep pace. As people increasingly delegate decisions to AI through everyday use — oversight through policy, regulation and organizational control is evolving more slowly. Organizations must therefore make their promises tangible by building systems that feel understandable, controllable and aligned with human needs. Trust can’t only live in principles, it has to show up in the experience of using AI.

Human-centered design has become one of the most powerful, yet under-used mechanisms that organizations have to close the trust gap.

As AI systems become more advanced, it's clear that building trust doesn’t happen through policies, principles or technical assurances alone. It is built, or lost, through experience. What people see, understand and feel when they use AI to make or influence decisions increasingly determines whether they are willing to try it again.

In this sense, design is becoming the front line of AI governance. It’s where real people experience safeguards, accountability and control. Thoughtful design becomes the bridge between capability and comfort. When designed well, AI provides reassurance and clarity. When designed poorly, it does the opposite.

“The most trusted experiences expand autonomy gradually,” explains Peter Neufeld, EY Studio+ EMEIA Financial Services Digital Customer Experience Leader, “giving users confidence first and then expanding what the system is allowed to do. Designing for comfort is about pacing — people don’t resist autonomy, they resist losing control.”

The EY Human Signals report (pdf) explores the key attributes of successful AI experiences:

• Emotional support: People looking for help care more about the emotion or complexity of the moment than the product they’re using.
• Simplicity: Good design reduces cognitive load by making an experience as effortless as possible — breaking down complex tasks and clarifying next steps.
• Proof and traceability: Superior design incorporates summaries, transcripts and logs so people feel like they are in control of the interaction, and to reduce misinterpretation.
• Adjustment: It gives people the control they want over how “warm” or “neutral” an AI tool feels. If needed, it provides instant access to interacting with a human instead of a bot.
• Inclusivity and accessibility: It has conversational interfaces that support people with different levels of literacy, language or accessibility.
• Visible constraints: Trust comes from clearly showing what the system will not do, and giving users control over those boundaries. Auditability, override and human escalation are not a backup plan but a core feature of accountability.

The future of AI is not being shaped by abstract debate or long‑range forecasts. It is being shaped by everyday choices — by when, where and how people decide to let AI act on their behalf.

The findings from this year’s survey make one thing clear: AI adoption is moving faster than sentiment. A growing minority of people is already delegating decisions to AI, while many more are building confidence through everyday, low‑risk uses. In some markets, this progression is already well underway, offering an early view of how quickly expectations can shift once familiarity and value align.

For business leaders, current market dynamics create both an opportunity and a responsibility.

The opportunity lies in recognizing that autonomy is not arriving all at once. It is expanding through specific moments, users and markets, often faster than organizations expect, but not without pattern or signal. The responsibility lies in understanding those signals early, learning from where confidence is already forming, and responding deliberately rather than reactively.

Crucially, this is not a choice between speed and caution. The data shows that people are not waiting for perfect trust before moving forward. They are negotiating confidence in motion — continuing to use and, in some cases, delegate to AI while simultaneously asking for clearer safeguards, stronger accountability and greater transparency.

It’s put leaders in a new position. The question is no longer whether autonomous AI will scale. It is whether organizations will shape scaling intentionally by accelerating where trust and value already exist, and slowing where clarity, safeguards or confidence are still needed.

What should leaders do next?

• Lead with experience, not promises: Build trust through use. Prioritize AI applications that deliver clear, everyday value and allow people to review, override or opt out as confidence grows.
• Segment by AI readiness, not demographics: People and markets are progressing at different speeds. Design experiences, messaging and controls that meet users where they are, rather than assuming a single path to adoption.
• Make trust visible: Providing clear disclosures, strong data protections, human‑in‑the‑loop options, and transparent accountability is no longer a differentiator. It is a prerequisite as autonomy increases.
• Design for emotional context: As AI takes on higher‑stakes tasks, design that considers empathy, simplicity and pacing is as important as performance. Design is not a finishing touch; it is a strategic catalyst for trust and adoption.

Leaders have a choice. They can allow delegation to expand by default, shaped by external forces and early adopters. Or they can design it deliberately, with visible constraints, clear accountability and confidence built in from the start.

Those who choose to act deliberately will do more than keep pace with change. They will earn a durable license to lead in an increasingly autonomous world.

The authors would like to thank Katherine Boiciuc, Oceania Chief Technology and Innovation Officer, Ernst & Young Australia; Peter Neufeld, EY Studio+ EMEIA Financial Services Digital Customer Experience Leader, Ernst & Young LLP, Krista Walpole, Associate Director, EYGS LLP; AnnMarie Pino, Associate Director, Ernst & Young LLP; Gaurav Batra, Associate Director, Ernst & Young LLP and Harshil Milan Zatakia, Supervising Associate, Ernst & Young LLP for their contributions to this article.

Summary

Everyday behavior shows that AI adoption is advancing faster than public confidence. While people often voice concern about trust and control, they continue to integrate AI into daily decisions, gradually relying on it for more consequential tasks. Familiar, low-effort uses are normalizing AI's presence and laying the foundation for greater delegation. As a result, decision-making authority is beginning to shift from people to systems, often before organizations have fully understood where acceptance is forming or how quickly expectations are changing.