AI Agent Autonomy, Escalation & Control Preferences Survey

Welcome! This survey explores your preferences for how AI agents should behave, when they should ask for your approval, and how they should hand off tasks to humans. Your participation is voluntary, and you may stop at any time. There are no right or wrong answers—we are interested in your honest opinions. All responses are confidential and will be reported only in aggregate. Estimated time: 11 minutes.

In the past 30 days, how often have you used any AI assistant (e.g., ChatGPT, Copilot, Gemini, Siri, Alexa)?

• Daily
• A few times a week
• About once a week
• A few times this month
• I have not used an AI assistant

The next few questions ask about specific behaviors an AI agent might exhibit. For each, please rate how important that behavior is to you.

By default, how much autonomy should an AI agent have when performing tasks in your workflow?

What concerns, if any, do you have about AI agents? (Select all that apply.)

• Privacy or data leakage
• Inaccuracies (hallucinations)
• Bias or unfair outcomes
• Security vulnerabilities
• Compliance or legal risk
• Unexpected costs or overuse
• Loss of personal skill or over-reliance
• Other (please specify)
• None of the above

Overall, how much do you trust AI agents to carry out tasks on your behalf?

Based on your responses in this survey, is there anything else you'd like to share about your expectations or preferences for AI agents?

Finally, a few optional questions about you to help us understand patterns across different groups.

Thank you for completing this survey. Your responses are confidential and will be used in aggregate to help improve AI agent design for everyone.

How important is it that an AI agent explains its reasoning before taking action?

When an AI agent is unsure about the correct action or detects a potential risk, what should it do first?

• Pause and escalate to a human immediately
• Ask me to review and approve the specific action
• Flag the uncertainty but continue with its best judgment
• Continue autonomously unless I intervene
• Other (please specify)

We'd like to explore your experiences with AI agents in a bit more depth. An AI moderator will ask you a couple of follow-up questions about situations where an AI agent met or fell short of your expectations.

What is your age range?

• 18–24
• 25–34
• 35–44
• 45–54
• 55–64
• 65+
• Prefer not to say

How important is it that an AI agent asks for your confirmation before performing irreversible actions?

If an AI agent escalates an issue to a human, which handoff method(s) would you prefer? (Select all that apply.)

• Live chat with a human agent
• Open a support ticket
• Email a specialist
• Phone call
• Schedule a callback
• Forward to my team or shared inbox
• Other (please specify)

In which region do you primarily live?

• Africa
• Asia
• Europe
• Latin America & Caribbean
• Middle East
• North America
• Oceania
• Prefer not to say

How important is it that an AI agent cites sources or evidence for its claims?

What is the maximum amount you would allow an AI agent to spend on a single purchase without your approval?

• $0 – always require my approval
• $1–$25
• $26–$100
• $101–$500
• $501–$1,000
• More than $1,000
• Not applicable – my agent would not make purchases

What is the highest level of education you have completed?

• High school or equivalent
• Some college or vocational training
• Bachelor's degree
• Master's degree
• Doctorate
• Prefer not to say

How important is it that an AI agent learns and improves from your corrections over time?

During longer tasks, how often should the AI agent send you status updates?

• Only when the task is complete
• At major milestones or decision points
• At regular intervals (e.g., every few minutes)
• After every individual step
• I don't need status updates

What is your current employment status?

• Full-time
• Part-time
• Self-employed
• Student
• Not employed
• Retired
• Prefer not to say

How important is it that an AI agent stays strictly within the scope of tasks you have assigned?

Which of the following actions, if any, should an AI agent be allowed to perform without asking you first? (Select all that apply.)

• Draft but do not send messages
• Send emails to internal recipients
• Schedule calendar events
• Browse and read web pages
• Access non-sensitive files
• Run code in a sandbox
• Create tickets or tasks
• Other (please specify)
• None of the above