AI Tools for Early Autism Detection Raise Ethical Concerns, Study Finds

Artificial intelligence-based clinical prediction models that mine electronic health record data could help identify autism earlier in childhood, but a new US-based qualitative study suggests that caregivers, clinicians, and autistic adults want such tools deployed with caution. The research, published in BMC Medical Ethics, examined the ethical questions surrounding these models—particularly around consent, transparency, equity, and how results are communicated to families.

The study involved semi-structured interviews and focus groups with three groups: eight caregivers of young autistic children, ten clinicians involved in screening and early intervention, and eight autistic adults. The researchers used thematic analysis to identify participants’ concerns, the sources of trust they described, and their preferences for how a child’s autism likelihood should be communicated within shared decision-making contexts. Because the study was qualitative and drew on a small sample, the findings reflect the perspectives of these particular participants rather than measurable outcomes of any deployed prediction tool.

Overall, participants expressed cautious optimism. They saw potential benefits in earlier identification of autism, a clearer understanding of developmental concerns, and more timely access to services. At the same time, they highlighted several risks. These included inequities that could arise from incomplete electronic health record data, the limited transparency of so-called black-box models whose internal reasoning is difficult to inspect, the emotional burden that very early screening could place on families, and the possibility that caregivers would be sidelined when tools operate without their input.

Timing emerged as a central theme. Participants felt screening was most valuable when conducted early—but not so early that families receive results without clear next steps. In their view, information about a child’s likelihood of autism is only meaningful when it is actionable, allowing families and clinicians to plan concrete follow-up rather than leaving them with uncertainty and no pathway forward.

The authors conclude that ethical implementation of AI-based clinical prediction models must extend beyond technical performance. They argue it should include embedded ethical practices that promote trust, equity, and supportive caregiver-clinician partnerships, and that the perspectives of interest-holders—caregivers, autistic adults, and providers—can inform how that implementation is achieved. As health systems consider adopting such models, the study suggests that questions of consent, communication, and equity deserve as much attention as predictive accuracy.

Subject of Research: Ethical considerations of AI-based clinical prediction models for early autism detection

Article Title: AI–based clinical prediction models for early autism detection: ethical considerations and use implications across caregivers, autistic adults, and clinical providers

Article References: Sperling, J., Haseley, E., Quenstedt, S., Jones, L. D., Brandsen, S., Brown, A., Muhigaba, P. B., Dawson, G., Engelhard, M., Maslow, G., Goldstein, B. A., & Franz, L. (2026). AI–based clinical prediction models for early autism detection: ethical considerations and use implications across caregivers, autistic adults, and clinical providers. BMC Medical Ethics. https://doi.org/10.1186/s12910-026-01634-w

Image Credits: AI Generated

DOI: 10.1186/s12910-026-01634-w

Keywords: artificial intelligence, autism, clinical prediction models, electronic health records, medical ethics, qualitative research, screening, shared decision-making, health equity, early detection, based, clinical