HKUST's ASD Super Agent Hits 0.92 AUC in Early Autism Screening

The Hong Kong University of Science and Technology (HKUST) announced ASD Super Agent on November 11, 2026, an AI platform for autism screening and family support that achieved a preliminary area under the curve (AUC) score of 0.92 — a statistical measure of how well a model distinguishes between children at higher and lower risk. The figure is preliminary. HKUST states that further clinical studies and real-world validation are planned before the platform is deployed.

HKUST describes the system as Asia's first integrated AI platform for autism support. It combines three functions that families normally access separately: early risk screening, personalized intervention recommendations, and continuous follow-up for parents and caregivers.

The stakes are concrete. According to HKUST, autism spectrum disorder affects approximately one in 100 children worldwide, while lengthy assessment waits, high treatment costs, and fragmented intervention information can prevent families from obtaining timely support. The platform is designed to assist families from the initial risk assessment, through home-based training recommendations, to longer-term follow-up.

Who built it, and on what evidence?

Kani Chen, a professor in HKUST's Department of Mathematics and Department of Industrial Engineering and Decision Analytics, leads the project. The development team grounded its work in field research rather than lab assumptions: it visited special education institutions, child development centers, and psychiatric hospitals in Hong Kong, the Greater Bay Area, and Beijing, and engaged with nearly 500 parents of children with autism. The researchers used that engagement to shape a practical support tool around the difficulties families actually reported.

That number matters for how the platform was designed. A tool built from close to 500 parent conversations reflects reported pain points — long waits, unclear guidance, scattered information — rather than purely clinical definitions of need.

How does the screening work?

Parents record short videos of a child's everyday activities with a smartphone or another smart device, complete standardized questionnaires, and provide basic developmental information. ASD Super Agent uses those inputs to generate a risk assessment and an analytical report. HKUST is explicit about the report's limits: it is intended as a reference when families seek professional evaluation, not as a replacement for that evaluation.

The privacy design follows a data-minimization approach, which means collecting and processing only the information described as essential for the analysis. The system prioritizes extracting posture features on the user's device itself. This matters because the screening inputs include home videos and children's developmental information — among the most sensitive categories of personal data, particularly when children are involved.

What does the intervention side offer?

Beyond screening, parents can describe specific challenges — emotional regulation difficulties, limited eye contact, communication difficulties, or repetitive behaviors — and receive customized support suggestions. Recommendations can include academic literature and background references, which HKUST intends to inform discussions between families and healthcare or educational professionals.

The platform's knowledge structure is what HKUST calls the 20+1 framework. It incorporates more than 20 autism intervention models, and it allows therapists, special education providers, and organizations to add their own methods and practical experience to the knowledge base. Named approaches include:

• Applied Behaviour Analysis (ABA), a widely used behavioral intervention method
• Naturalistic Developmental Behavioural Interventions (NDBI), which embeds teaching in everyday play and interaction
• TEACCH Structured Teaching, an approach that organizes the physical environment and routines
• The Localized BSR Model

This structure combines research-based methodologies with frontline professional experience. It also gives professionals a route to contribute their own intervention knowledge, rather than limiting the system to the initial set of more than 20 models.

How does long-term follow-up work?

For continuing support, the system combines text records, video assessments, and developmental information supplied by families and professionals into a dynamic developmental profile for each child. The system can divide long-term goals into manageable daily actions, track progress over time, and support discussions between families and professionals. Voice interactions are currently supported in Cantonese, Putonghua, and English — a language set aligned with the platform's target regions of Hong Kong and the Greater Bay Area.

The research foundation is substantial. The team has built a multimodal autism database containing more than 1.1 million video, posture, and behavioral records. HKUST says the database provides a foundation for AI model development and autism research. Continued data collection and analysis are intended to improve risk-assessment accuracy, refine intervention recommendations, and deepen understanding of autism development and support strategies.

What comes next?

The researchers plan to refine the user experience, expand clinical validation studies, and strengthen system performance, with the stated goal of making autism support more accessible in Hong Kong, the Greater Bay Area, and beyond.

The numbers to watch are already on the table. A preliminary AUC of 0.92, a database of more than 1.1 million records, and input from nearly 500 parents give the project an unusually concrete evidentiary starting point for an assistive tool in this domain. But HKUST itself frames the decisive test ahead: the central question for the next stage is whether clinical and real-world testing can substantiate the preliminary screening result and the platform's integrated approach to family support.