Why Public Health Leaders Must Own AI Risk Decisions Now

Artificial intelligence is no longer a distant prospect for public health agencies; it is already reshaping how officials detect outbreaks, allocate resources, and design prevention programs. Yet according to a new commentary published in the International Journal of Public Health, the conversation about AI in public health has been dominated by technical specialists and regulators, leaving a critical gap at the very top of organizations. Olusimbo Ige of the Chicago Department of Public Health and Dolapo Fakuade of Coventry University and Khazar University argue that managing AI risk must become a core executive responsibility, not a task delegated to data scientists.

Their commentary, published in October 2026, lays out a decision framework designed to help public health leaders interrogate AI systems before, during, and after deployment. The authors' central concern is the environment in which public health operates. Unlike a hospital treating a single patient, public health agencies work with incomplete information, under intense political scrutiny, and amid competing priorities shaped by complex social determinants of health.

In this setting, an algorithmic error is not merely a technical glitch; it can severely damage public trust in the institutions responsible for protecting entire communities. A biased prediction or an inaccurate outbreak forecast can ripple outward, affecting which neighborhoods receive services and which populations are overlooked. That is why the authors insist that AI be viewed not just as a technical tool but as a major component of executive decision-making, with leaders accountable for balancing technical evidence against human judgment.

The commentary traces how the nature of executive decisions has shifted. Historically, public health leaders bridged what the authors call the data-to-action gap by translating descriptive statistics, such as a 12 percent increase in overdose mortality, into strategic responses. Today, AI-enabled epidemiology can integrate surveillance data, intervention effectiveness, and geospatial patterns to deliver predictive and prescriptive intelligence directly to decision makers.

An AI tool can identify high-burden communities and estimate how scaling a specific outreach program would prevent a precise number of deaths. These predictive efficiencies are driving adoption across the field. But the authors caution that the assumptions and trade-offs connecting epidemiological observations to recommended actions often remain concealed, and transparency is essential because public health choices are rarely based solely on mathematical predictions; they demand consideration of ethics, equity, and political context.

To equip leaders for this shift, the commentary proposes a foundational expansion of executive competency: AI literacy. The authors stress that AI is not a single technology but a collection of distinct approaches with different functionalities and risk profiles. Machine learning identifies patterns within historical data to generate classifications.

Predictive AI forecasts future events such as disease outbreaks. Generative AI creates new content, including reports and text summaries. Computer vision processes visual data, such as radiological images, for automated surveillance.

Most recently, AI agents have emerged as autonomous systems capable of executing multi-step tasks, retrieving information across platforms, and coordinating workflows under human oversight. Because each of these carries different risks, the authors argue that AI literacy is no longer a niche technical skill but an essential core competency for contemporary public health leadership. One of the sharpest tensions the commentary identifies lies between individual-level and population-level AI.

Most commercial AI systems are built for precision medicine, predicting risk for individual patients in clinical settings. Public health, by contrast, focuses on collective risk identified through population-wide surveillance and prevention. The authors warn that AI built for hospitals is not automatically suited for public health workflows, and that deploying tools designed for the wrong paradigm, such as precision-medicine models applied to population surveillance, is a concrete failure mode executives must guard against.

Leaders must therefore ask whether a given tool is necessary and appropriate for the public health decision at hand, and whether its intended use aligns with a population-level prevention objective rather than an individual-level clinical prediction or treatment objective. Explainability emerges as another central leadership responsibility. Many high-performing models are difficult to explain to policymakers, practitioners, or affected communities, yet trust hinges on how clearly a system can account for itself.

The commentary offers a concrete illustration: an overdose prediction model trained primarily on insured populations will underperform in marginalized communities. Executives must enforce strict documentation and traceability, vetting what data was used, how models were built, how decisions were generated, and who approved deployment. The framework asks whether an organization can explain, in terms appropriate for decision makers and communities, how the system generates its recommendations, and whether there is documented traceability for data sources, model development or procurement, versioning, validation, deployment approvals, and changes made after launch.

Without this, the authors warn, agencies risk opaque black-box decisions that cannot withstand public or political scrutiny. The regulatory landscape adds another layer of complexity that the authors say executives cannot ignore. In 2026, more than 43 US states introduced AI bills targeting chatbot oversight and clinical disclosure requirements.

The Centers for Medicare and Medicaid Services opened rulemaking on AI-powered fraud detection, while the European Union AI Act established a risk-based framework that classifies most population-health AI as high-risk. Navigating this patchwork of state, federal, local, contractual, and international requirements requires strong governance, and the commentary stresses that data security, interoperability, and privacy compliance cannot simply be delegated to algorithms. Executives need a plan for tracking a fast-evolving policy landscape, because non-compliance carries both legal and reputational exposure for agencies already operating under public scrutiny.

Beyond compliance, the authors urge leaders to look past AI's efficiency and assess its acceptability, particularly for tools that affect entire populations. Executives must define what constitutes unacceptable risk before deployment. Following EU AI Act guidelines, AI must not discriminate or restrict access to services, which demands particular caution and governance when AI is used for risk prediction that determines resource allocation or program eligibility.

The commentary sketches a spectrum of risk: chatbots and virtual assistants can be lower risk when paired with transparency and defined data sources, while minimal-risk applications such as spam filters and scheduling tools carry few regulatory requirements. The point is that risk assessment must be proportionate and deliberate, calibrated to how consequential and how population-facing a given application actually is. Continuous human oversight forms the final safeguard in the framework.

The authors call for mandatory monitoring and evaluation mechanisms to assess safety, equity, and real-world performance against population health goals. Epidemiologists must review outbreak alerts, clinicians must validate high-risk recommendations, and leaders must retain final decision authority. Organizations must monitor model drift, accuracy declines, and algorithmic bias over time, because a model that predicts overdoses accurately today may fail if the demographics of the population shift.

Mitigating these risks requires multidisciplinary input from epidemiologists, data scientists, legal counsel, and equity experts, and AI recommendations should be validated against observed results so corrective action can be taken. The core principle, as the authors state it, is that AI should not be deployed until it has been comprehensively reviewed for vulnerabilities. The commentary concludes that public health decision-making is not merely technical.

Leaders do not need to know how code works, but they must be able to verify what data trained the AI, who is accountable when it errs, and whether the system can explain its outputs. Transparent, disciplined AI governance, they argue, is now the primary differentiator for modern public health organizations, and the framework offers executives a practical starting point for making informed decisions about AI-related risks before high-consequence choices are made on their watch.

Subject of Research: Executive-level frameworks for managing artificial intelligence risk in public health organizations

Article Title: Executive decision support for managing AI risk in public health

Article References: Ige, O., & Fakuade, D. (2026). Executive decision support for managing AI risk in public health. International Journal of Public Health, 71, Article 1610326. https://doi.org/10.3389/ijph.2026.1610326

Image Credits: AI Generated

DOI: 10.3389/ijph.2026.1610326

Keywords: artificial intelligence, public health, AI governance, executive decision-making, algorithmic bias, health equity, machine learning, predictive AI, generative AI, regulation, human oversight, public health leadership