AI Safety, Risk Perception & Boundary Behaviour 2025 (Dataset)

This dataset is part of the Human Clarity Institute’s AI–Human Experience Data Series. It measures how adults evaluate the safety of AI systems, perceive risk levels, identify boundary-crossing behaviours, and respond to uncertainty or potentially unsafe AI outputs. The dataset includes:• 1–7 Likert-scale items measuring AI safety concerns and risk perception• scenario-based boundary judgement tasks• indicators of precaution, caution, and willingness to disengage from AI systems• assessments of trust, uncertainty, and perceived overconfidence in AI responses• open-text reflections on uncomfortable or unsafe AI interactions• demographic variables across six English-speaking countries Data were collected via Prolific in 2025 from adults across the UK, US, Australia, Canada, New Zealand, and Ireland. All data were cleaned, anonymised, and verified according to the Human Clarity Institute’s open-data publication protocol. This dataset contributes to understanding how people perceive AI safety and risk in everyday settings, offering early insight into human behavioural thresholds, safety expectations, and boundary concerns as AI tools grow more capable.