Jul 9, 2026 UVA researchers say AI tools could pose risk to athlete health, military readiness
Researchers from the University of Virginia released a paper last month arguing that many artificial intelligence tools have not demonstrated the real-world effectiveness needed to justify their use and instead pose new risks from false results and other technical deficiencies in models hastily rushed to market, often by government agencies.
The paper was published in late June in “Medicine & Science in Sports & Exercise” and authored by researchers from UVA’s School of Data Science and School of Education and Human Development.
The authors conclude that many commercially available AI tools lack sufficient transparency, external validation and accountability, creating risks for athletes, patients and military personnel.
“Even if a coach is content with ‘what works’ in practice, our responsibility as data scientists and clinicians is to maintain an ecosystem of information that reflects underlying physiological or biomechanical processes,” the authors said in a news release. “When we base decisions on rigorously vetted casual relationships rather than on spurious associations, we create training and rehabilitation protocols that are both effective and safe.”
According to the authors, their findings are particularly significant for military readiness.
Musculoskeletal injuries are among the leading causes of lost readiness, disability and healthcare utilization across the military. The authors note that AI-driven systems are increasingly being used to identify service members at risk of injury, assess fitness, guide training programs and predict readiness outcomes. However, drawing on evaluations of AI-based systems deployed in elite military training environments, the researchers found little evidence that some commercially available tools performed better than chance when attempting to predict injuries among service members.
Key findings of the study include:
- Researchers found that several AI-driven injury prediction systems used in military settings showed poor predictive performance and limited reliability when tested in large cohorts of service members.
- The authors argue that many current systems operate as “black boxes,” making it difficult for clinicians, coaches and military leaders to understand how recommendations are generated or whether they are grounded in meaningful physiological mechanisms.
- The authors warn that inaccurate readiness or injury-risk assessments can have real-world consequences, including unnecessary restrictions on training, missed injury risks, disrupted mission preparation and reduced operational effectiveness.
Additionally, the authors call for stronger oversight of AI tools used to generate health, injury-risk and readiness recommendations.
“The FDA’s AI/ML-Based Software as a Medical Device Action Plan outlines expectations for algorithm transparency, continuous real-world performance monitoring and premarket validation to ensure clinical reliability,” they said. “Applying similar regulatory principles to sports science software — particularly those that generate health, injury-risk or readiness outputs — would add essential rigor and accountability.”
Despite the increasing use of wearable technologies and AI-driven analytics by professional sports and military organizations, the authors note that independent review of the algorithms behind these systems is still limited.
“For instance, FIF’s Quality Programme rigorously tests wearable and tracking equipment for basic data-collection accuracy prior to club adoption, though it currently stops short of assessing the proprietary predictive models that accompany these tools,” they said. “While other professional sports leagues reportedly maintain similar review processes, their findings remain unpublished, dampening both transparency and competitive pressure on vendors to uphold best practices.”
The researchers emphasized that they are not dismissing the future potential of AI in sports medicine, healthcare or military performance but instead recommended independent external validation, adversarial testing, ongoing performance monitoring and greater transparency before AI systems are widely integrated.
“These methodologies themselves hold considerable promise,” they said. “But premature commercialization without rigorous validation has eroded confidence and slowed progress.”
