Artificial intelligence is moving toward one of the most consequential—and least examined—frontiers in health care: correctional systems. A Viewpoint published in JAMA Internal Medicine examines how AI could help address serious gaps in medical care for incarcerated people while also creating new mechanisms for discrimination, surveillance, and neglect. The authors argue that algorithms cannot be treated as neutral tools in prisons and jails, where patients often have complex medical needs, limited control over their health information, and fewer opportunities to challenge clinical decisions.
The article, led by Lawrence A. Haber, MD, of Denver Health and Hospital Authority, focuses on the tension between AI’s technical capabilities and the structural conditions of correctional medicine. Artificial intelligence can analyze large volumes of health records, identify patterns that clinicians might miss, and support decision-making in settings where staffing shortages and restricted access to specialists are common. In theory, these systems could help detect deteriorating health, prioritize urgent cases, improve medication management, and connect incarcerated patients with remote physicians through telemedicine.
Yet the same systems could deepen inequities if they are trained on incomplete, biased, or unrepresentative data. Machine-learning models generally identify statistical relationships in historical information rather than understand the causes of illness or the social circumstances surrounding a patient. If past records reflect delayed diagnoses, limited treatment, or unequal access to care, an algorithm may reproduce those patterns while presenting them as objective predictions. A model designed to estimate the likelihood of hospitalization, for example, might systematically underestimate the needs of people whose symptoms were previously overlooked or poorly documented.
Correctional health systems pose distinctive challenges for the development and use of these technologies. Medical records may be fragmented across jails, prisons, hospitals, and state agencies, making it difficult to assemble accurate longitudinal data. Documentation can also be shaped by security priorities rather than clinical needs. Information about mental illness, substance use, disability, infectious disease, or pregnancy may be incomplete, inconsistently coded, or inaccessible to the clinicians responsible for care. An AI system working with such data could generate confident recommendations that are nevertheless based on a distorted picture of a patient’s health.
The consequences of algorithmic error may be particularly severe behind bars. Incarcerated patients cannot freely seek a second opinion, change clinicians, or leave a facility to obtain alternative care. A risk score could influence whether someone receives an urgent evaluation, is transferred to a hospital, or is placed under heightened observation. If staff treat an automated output as a final answer rather than one piece of evidence, a technical error could become a medical decision with little transparency or recourse. The authors therefore emphasize that AI should support—not replace—professional judgment and direct communication with patients.
Privacy is another central concern. Health data used by AI systems can include diagnoses, laboratory results, psychiatric histories, genetic information, medication records, and behavioral observations. In correctional settings, the boundary between medical information and institutional surveillance is often difficult to maintain. Data gathered for clinical purposes could be repurposed to monitor behavior, assess disciplinary risk, or make administrative decisions unrelated to health. The Viewpoint calls for clear limits on data collection, access, retention, and secondary use, along with safeguards that preserve the confidentiality essential to effective medical treatment.
Equity also depends on whether incarcerated people are meaningfully included in decisions about AI. Patients should be informed when algorithmic tools influence their care and should have a way to question or appeal consequential decisions. Independent oversight could examine whether models perform differently across racial, ethnic, gender, age, disability, and language groups. Technical evaluation should include more than overall accuracy: developers and health systems should measure false-negative and false-positive rates, calibration, interpretability, and clinical outcomes for groups that have historically experienced inadequate care.
The authors further point to the importance of governance before deployment. Correctional agencies and health care providers would need procedures for validating algorithms in real-world facilities, monitoring performance over time, documenting errors, and suspending systems that produce harmful results. Clinicians should receive training in the limitations of predictive models, including the danger of automation bias—the tendency to accept a computer-generated recommendation even when professional judgment raises doubts. Procurement contracts should require transparency about training data, model updates, security practices, and vendor responsibility when systems fail.
The larger message is that AI cannot solve inequity without confronting the institutional conditions that create it. Faster triage or more sophisticated prediction will not compensate for inadequate staffing, poor infrastructure, restricted specialty care, or policies that treat incarcerated patients as less deserving of medical attention. Used cautiously and governed rigorously, AI may help correctional health systems identify unmet needs and extend clinical capacity. Used without accountability, it could make unequal care more efficient and harder to challenge. The Viewpoint urges policymakers to make health equity, patient autonomy, privacy, and human oversight central requirements before artificial intelligence becomes embedded in the medical routines of prisons and jails.
Subject of Research: The potential benefits and risks of artificial intelligence in correctional health systems, with a focus on health inequities among incarcerated populations.
Web References: https://doi.org/10.1001/jamainternmed.2026.3471
References: Haber LA. Viewpoint on artificial intelligence, correctional health systems, and health equity. JAMA Internal Medicine. doi:10.1001/jamainternmed.2026.3471.
Keywords: Artificial intelligence; internal medicine; imprisonment; health care; health inequities; incarcerated populations; health care policy.
