Patient Redress for AI-Related Medical Errors in Bangladesh: A Socio-Legal Survey

Responsibility for artificial intelligence (AI) assisted medical decisions encompasses both preventing harm and providing a way to respond when harm occurs. These functions are often combined in empirical measures of accountability, making it difficult to identify which safeguards respondents prioritise. This cross-sectional analysis used 385 adult responses to a common questionnaire on AI, health law and patient rights in Bangladesh. Three preventive statements addressed defined responsibility, independent clinical judgement and organisational and developer safety controls. A separate statement addressed access to complaints and remedies; a fifth elicited anticipated trust under clear accountability. Endorsement was compared within respondents using exact McNemar tests with Holm adjustment. Exploratory regression distinguished prevention and redress as correlates of anticipated trust. Redress was endorsed by 311 respondents (80.8%), exceeding endorsement of each preventive statement by 16.6–20.5 percentage points (all adjusted p < 0.001). The adjusted linear association with anticipated trust was positive for the prevention summary (b = 0.205, 95% CI 0.049 to 0.361; adjusted p = 0.020), but attenuated in inference when trust was dichotomised. The redress coefficient was close to zero in the adjusted linear model (b = 0.001, 95% CI −0.112 to 0.114). These findings distinguish broad endorsement of a patient-facing remedy from covariance with a hypothetical trust judgement. They support evaluating complaint access as a governance outcome in its own right. The survey neither verifies actual use of medical AI nor estimates effects of introducing safeguards.