Frequency was not a simple proxy for weaker agency
How people engaged with AI appeared more informative than how often they used it.
Neuravox Policy Lab
Research report · August 2026
Human Agency in AI Use.
Exploratory findings from surveyed generative AI users
- Programme
- Neuravox Policy Lab
- Author
- Gideon Abako
- Publisher
- Neuravox Foundation
About the report
How people retain control while using generative AI
Human Agency in AI Use examines how everyday generative AI use relates to independent thinking, verification, explanation, capability retention and decision authority. The exploratory study combines primary survey evidence with research and policy literature on cognitive offloading, automation bias, critical thinking, AI literacy and human autonomy.
The report is the first evidence-building step in a wider Neuravox programme to develop and test a practical Human Agency Framework for everyday AI use.
Exploratory findings
Three patterns for further study
Retained control coexisted with delegation
AI assistance and behaviours associated with verification, explanation and independent judgment appeared together.
Agency appears multidimensional
The evidence did not support compressing the surveyed behaviours into one overall agency score.
Emerging framework
Six working domains of human agency
The report proposes these domains as a working structure for later qualitative research, improved measurement and validation. They are not presented as a final or validated scale.
Independent initiation
Beginning a task with an independent view before turning to AI.
Cognitive authorship
Remaining actively responsible for the structure and direction of ideas.
Verification and epistemic control
Checking information and challenging responses that conflict with judgment.
Explanatory ownership
Understanding and being able to explain the final work.
Capability retention
Retaining the ability to complete similar tasks without AI.
Decisional authority
Keeping responsibility for consequential choices with the person.
Implications
From usage counts to the quality of human control
Individuals
Use AI in ways that preserve independent thought, verification, explanation and the ability to act without the system.
Organizations
Evaluate the quality of human oversight and retained capability, not only the intensity of AI use.
AI designers
Build interactions that support reflection, challenge, learning and meaningful human decision authority.
Policy
Translate broad commitments to human oversight into observable practices for everyday AI use.
Suggested citation
Abako, G. L. (2026). Human Agency in AI Use: Exploratory findings from surveyed generative AI users. Neuravox Policy Lab, Neuravox Foundation.