ntroduction

Artificial Intelligence (AI) is rapidly reshaping public health — from enhancing disease surveillance and diagnostics to easing workforce burdens — but it also raises complex risks and ethical questions. In Europe and globally, public health leaders are grappling with how best to harness AI’s revolutionary potential while managing its pitfalls. After decades of experience, many recognise that AI is not a magic fix for health challenges; its value depends on thoughtful integration into health systems. This article provides an in-depth review of the current relationship between AI and public health. It examines the opportunities it offers, real-world innovations already underway, practical implementation challenges, and the risks and governance frameworks that must guide responsible use. All discussions equally consider European contexts (including emerging EU regulations) and broader global health perspectives.

TL;DR Summary

• AI’s growing role in health: Artificial intelligence is increasingly used to augment public health efforts — from automating administrative tasks to advanced disease surveillance and diagnostics — offering new ways to improve efficiency and reach.
• Tangible benefits observed: Early deployments show promising results. AI tools have reduced clinicians’ paperwork burden, flagged outbreaks days before traditional systems, and enhanced diagnosis in low-resource settings (e.g. catching 15% more TB cases via X-ray analysis).
• Innovations across sectors: NGOs, governments, and companies are all investing in AI for health. For example, PATH and others use AI in field programmes, the NHS has dozens of AI pilots improving care delivery, and pharma companies leverage AI to speed up drug and vaccine development.
• Practical hurdles remain: Successful implementation requires robust data infrastructure, interoperability, and high-quality data. Many health systems must modernise IT systems and address data silos and quality issues before AI can perform optimally.
• Human factors are critical: Integrating AI into workflows and gaining staff acceptance are significant challenges. Training health workers, providing explainable outputs, and maintaining human oversight are essential to building trust in AI-assisted care.
• Key risks to manage: AI in public health brings serious risks — privacy breaches, algorithmic bias harming disadvantaged groups, opaque “black box” decisions undermining trust, and AI-generated misinformation spreading false health advice. Over-reliance on AI without safeguards can also be dangerous.
• Ethics and governance frameworks: Clear principles and regulations are emerging to guide responsible AI use. WHO’s six ethical principles (e.g. transparency, equity, accountability) set value-based guardrails, while the EU’s AI Act will enforce strict requirements on high-risk health AI (mandating transparency, risk management, and human oversight).
• Collaboration and capacity-building: Effectively advancing AI in public health will require interdisciplinary collaboration (health experts with technologists), investment in workforce AI literacy, and inclusive approaches that involve LMICs and marginalised groups so benefits are shared widely.
• Continuous evaluation and adaptation: To ensure AI delivers on its promise, public health authorities must continually monitor outcomes, audit algorithms for bias or errors, and be ready to adjust or suspend systems if problems arise. Adaptive governance and ongoing community feedback are vital for safe, effective AI integration.
• Seizing the opportunity responsibly: When guided by ethical principles and strong oversight, AI can greatly strengthen public health, easing workforce burdens, expanding outreach, and providing data-driven insights. The next few years are crucial for implementing the policies, education, and trust-building measures that will allow AI to be a force for health equity and innovation rather than a source of new disparities or dangers.

Opportunities: Transforming Public Health with AI

AI is being deployed to alleviate several longstanding public health challenges. One significant opportunity is reducing clinician burnout and workforce shortages by automating routine tasks. For example, a 2024 survey found that 57% of physicians believe automating administrative burdens is the top opportunity for AI to ease workloads amid staff shortages. Machine learning systems can transcribe medical notes, pull up patient records, and handle scheduling or prescription refills — freeing clinicians to spend more time on patient care. Many doctors see such automation as a key to improving work efficiency and reducing stress, suggesting AI could help mitigate the healthcare burnout epidemic.

AI also offers powerful tools for disease surveillance and epidemic intelligence. Algorithms can continuously scan vast data sources — news reports, social media, travel data — to spot early signs of outbreaks far faster than traditional methods. Notably, the HealthMap and BlueDot platforms (which use natural language processing and machine learning) flagged the COVID-19 outbreak days before official alerts. By sifting through informal signals and anomalies, AI-driven systems can provide precious early warnings of emerging health threats. BlueDot’s AI surveillance tools have dramatically sped up outbreak detection, reducing manual scanning time by nearly 90% in some cases. Such early alerts enable public health agencies to mobilise quicker responses and potentially contain outbreaks before they spread.

Another area of opportunity is improving diagnostics and clinical decision support, especially in resource-constrained settings. AI image recognition has shown great promise in interpreting medical images like X-rays and retinal scans. For example, AI-based chest X-ray tools for tuberculosis (TB) are being used to help screen patients in low-resource areas that lack radiologists. A recent programme in India led by PATH found that an AI tool (qXR) boosted TB case detection by ~15.8% — identifying cases that human readers missed. Many countries are now utilising AI-assisted chest X-ray screening for TB, which can lead to earlier diagnosis and treatment in underserved communities. Beyond imaging, AI-powered diagnostic apps and chatbots can guide patients through symptom checks or flag high-risk cases for follow-up, expanding access to essential healthcare advice where clinicians are scarce.

Crucially, AI is also being enlisted to address climate-related health threats and environmental impacts on health. Public health researchers increasingly pair AI with climate data to predict disease patterns under changing environmental conditions. For instance, machine learning models can correlate weather patterns (temperature, rainfall) and even animal health data with disease outbreaks to anticipate risks in specific locations. By analysing such data, AI-driven predictive analytics can serve as early warning systems — forecasting surges in vector-borne diseases like malaria following heavy rains or heat-related illness during extreme heatwaves. This capability is ever more critical as climate change intensifies health hazards. AI can help public health officials prepare for climate-sensitive disease outbreaks, allocate resources proactively, and develop adaptation strategies to protect vulnerable populations.

Real-world Applications and Innovations

AI in public health is not just theoretical — numerous real-world initiatives by NGOs, governments, and private companies have already demonstrated its potential. Global health nonprofits and international agencies have been early adopters of AI to support their missions. For example, the Bill & Melinda Gates Foundation has invested heavily in AI-driven global health projects. In 2023, it awarded grants to nearly 50 pilot projects exploring AI solutions for health and development challenges — these range from AI-augmented diagnostic tools to data systems for disease surveillance in low-income settings.

One Gates-backed innovation is AI-assisted ultrasound: in 2020, a $44 million grant was given to develop an AI-guided portable ultrasound to improve lung disease diagnosis in low-resource countries (e.g. detecting pneumonia). Likewise, PATH and other NGOs are integrating AI into field programmes — as seen in the TB screening project, where an AI tool significantly increased case finding while illuminating practical deployment hurdles. These efforts by NGOs underscore AI’s promise to close gaps in healthcare access and quality for underserved populations.

Governments and public health agencies are also launching AI initiatives. In Europe, national health systems pilot AI to improve services and efficiency. For instance, the UK’s National Health Service (NHS) created an NHS AI Lab to fund and evaluate AI innovations in care delivery. By 2025, the NHS had over 80 AI projects live, targeting everything from optimising nurse rostering and predicting hospital bed occupancy to speeding up radiology workflows.

One NHS program provided £100+ million in awards to develop AI for earlier cancer detection, resource management, and patient safety improvements. The NHS AI Lab’s “Skunkworks” team has run short-term projects that yielded practical tools — e.g. an algorithm to streamline the placement of nurses across wards and a natural language processing engine to search health records more efficiently. Meanwhile, European public health agencies are leveraging AI for epidemiology; the European Centre for Disease Prevention and Control (ECDC) has incorporated systems like BlueDot’s AI to enhance epidemic intelligence, including monitoring outbreaks during events such as the 2020 Olympics. These government-led efforts illustrate growing public sector commitment to deploying AI for health system strengthening and emergency preparedness.

The private sector, particularly in healthcare and pharmaceuticals, is likewise driving innovation at the intersection of AI and public health. Pharmaceutical companies now routinely use AI in drug discovery and development. For example, Novartis recently struck a wide-ranging partnership (worth up to $1 billion) to use a generative AI platform for designing new protein-based therapies — aiming to accelerate the search for novel disease treatments. GSK has also embraced AI to speed up R&D: its CEO noted that AI modelling helped cut two years off an RSV vaccine trial by predicting where outbreaks would occur and optimising trial site selection. This led to the faster development of the world’s first RSV vaccine, an essential public health breakthrough.

Beyond pharma, medical technology firms are integrating AI into devices, from smart wearables that flag irregular heart rhythms to imaging systems where AI assists in analysing scans for early signs of cancer. Startups and tech companies are introducing AI-driven health apps and chatbots (such as symptom checkers and mental health conversational agents), which some health services in Europe are trialling for patient triage and support. These real-world examples underscore that AI is already deeply enmeshed in the health ecosystem — from global disease surveillance networks to hospital wards and R&D labs — delivering innovations that could improve population health outcomes.

Practicalities and Implementation Challenges

While the potential is immense, implementing AI in public health is a pragmatic challenge. Infrastructure and data interoperability are foundational hurdles. Effective AI requires robust digital infrastructure — high-quality data streams, electronic health records, and cloud computing capacity — which many health systems lack, especially in low-resource settings. Data needed for public health AI often reside in silos or incompatible formats across hospitals, labs, and agencies. Poor interoperability means AI tools struggle to aggregate and interpret information from disparate sources. Bridging these gaps will require significant investment in health information systems, common data standards, and connectivity. Encouragingly, current AI technology can assist in standardising and mapping messy health datasets to make them more usable. Nonetheless, without reliable infrastructure and data-sharing frameworks, even the best AI algorithms cannot deliver consistent results across a public health network.

A related challenge is data quality and representativeness. AI models are only as good as the data they learn from, and health data can be incomplete, biased, or unrepresentative of specific populations. Studies highlight issues like variability in how data are recorded, large amounts of unstructured text, missing information, and coverage bias (e.g. most training data coming from high-income populations).

These factors can undermine an AI system’s accuracy and value to end users. Developing good AI for health requires carefully cleaning and curating data to reflect clinical reality. For instance, algorithms trained only on European hospital data may perform poorly in rural African communities. Implementers must thus invest effort in data preparation and continuously monitor model outputs for anomalies. Establishing metadata standards, common terminologies, and data quality metrics can facilitate better AI development. Additionally, clarity on data ownership and governance is needed: questions about who “owns” health data (patients, providers, governments?) affect how data can be integrated for AI. Resolving these issues through policies and trust frameworks is key to unlocking data for public health AI while respecting privacy and rights.

Another practical consideration is integrating AI tools into healthcare workflows and gaining workforce acceptance. Introducing AI decision-support systems or automation in clinics requires adapting processes and training staff. Health workers may be understandably cautious — some lack familiarity with AI, worry about accuracy, or fear being displaced. Clear protocols are needed if an AI system’s recommendation conflicts with clinical judgment. Early experience shows that human-AI collaboration works best when AI is framed as an assistive tool rather than a professional replacement. Building trust among the workforce involves providing explainable outputs and demonstrating reliability in pilot phases. It also means training clinicians in basic AI concepts and ensuring they feel confident interpreting AI outputs.

Successful deployments (like the PATH TB screening program) emphasise that significant workflow integration and training efforts are required. In that program, implementers had to solve issues of installing the software in clinics, securing internet connectivity for the AI, and ensuring staff could effectively use the AI results within their screening workflow. Without such groundwork, even a high-performing algorithm might sit on the shelf unused. Thus, the human element is crucial: public health organisations must engage and educate their workforce, adjusting roles and processes so that AI enhances rather than disrupts care delivery. Over time, as clinicians see AI reducing drudgery (e.g. auto-filling forms) and improving outcomes, their acceptance tends to grow. Indeed, physician enthusiasm for health AI has been rising year-on-year. Patience and iterative refinement are needed to blend AI smoothly into the complex fabric of health systems.

Risks and Concerns of AI in Public Health

Despite the optimism, it is vital to acknowledge the risks and potential harms associated with AI in public health. Data privacy and security tops the list of concerns. AI systems often require large datasets of patient information, raising the stakes for protecting sensitive personal health data. Any breach or misuse of such data can erode public trust and violate individuals’ rights. There is also the risk of “function creep”, where data collected for health purposes might be used in other ways (for example, a COVID-19 contact tracing app’s data later being used for law enforcement — a scenario that drew criticism in some countries). Moreover, complex AI models could inadvertently leak private details — for instance, a model might be reverse-engineered to reveal records it was trained on. Ensuring robust cybersecurity and strict data governance is therefore paramount. Many call for comprehensive privacy safeguards and compliance with regulations like Europe’s GDPR whenever AI handles health data. Techniques such as anonymisation or synthetic data can help, but they are not foolproof (even de-identified data can sometimes be unidentified).

The bottom line: without public confidence that AI will maintain confidentiality and data security, its benefits will be lost. Public health agencies must be transparent about what data are used and how to obtain informed consent where appropriate and implement state-of-the-art security measures to prevent breaches. Privacy isn’t just a legal box to tick — it’s fundamental to preserving the trust on which public health interventions depend.

Another significant risk is algorithmic bias and the exacerbation of health inequalities. AI systems can unintentionally perpetuate or even worsen disparities if their design is not carefully managed. This was starkly illustrated by a widely used healthcare risk algorithm in the United States that was found to be racially biased. The algorithm helped determine access to extra care programs and used healthcare cost as a proxy for need. This choice systematically underestimated the needs of Black patients (who often had lower healthcare expenditures due to access barriers). As a result, many high-risk Black patients were less likely to be flagged for additional care, denying them the resources they needed. This example shows how bias in data or design can translate into inequitable outcomes: the AI effectively discriminates against a vulnerable group. Similar issues could arise in public health if an AI model is trained on predominantly male patients under-detect conditions in women or if disease surveillance AI better covers wealthier communities with more data. AI could widen gaps if not addressed, with marginalised populations benefiting the least or even being harmed.

Equity must be a central design principle to counter this: datasets should be diverse and inclusive, algorithms should be tested for bias, and bias mitigation strategies (like reweighing data or algorithmic fairness adjustments) should be applied. The WHO explicitly highlights inclusiveness and equity as core ethical principles for AI, ensuring that AI tools work for all segments of society regardless of race, gender, income, or other characteristics. Ultimately, careful governance and auditing of AI systems are needed to avoid encoding systemic biases into digital form and instead use AI to reduce health inequities (for example, by targeting interventions to underserved areas).

A further concern is the lack of transparency (“black box” issue) and its impact on trust and safety. Many AI models, especially deep learning networks, operate as complex black boxes — they do not explain their reasoning in human-understandable terms. In healthcare, this opacity is problematic. Clinicians and public health decision-makers are wary of acting based on a recommendation they don’t understand, particularly if an AI’s advice contradicts intuition or standard practice. Unexplainable AI can also undermine accountability: if an AI makes a harmful mistake, it may be unclear why it happened or who is responsible. This lack of transparency feeds directly into trust issues among professionals and the public. If people perceive AI as a mysterious, untrustworthy “magic wand” imposed on health decisions, they may reject its use. There have been cautionary tales: an AI system deployed in hospitals to predict which COVID-19 patients would need ICU care was later found to underperform because it hadn’t been adequately validated. Clinicians grew sceptical of its risk scores.

To prevent such scenarios, experts call for explainable and interpretable AI in health — algorithms that can provide reasons for their predictions or use transparent, logical rules where possible. At a minimum, users should have access to information about how an AI was developed and its known limitations. Regulatory frameworks like the EU AI Act are likely to mandate a degree of transparency for high-risk AI (including many medical applications) precisely to bolster trust and enable oversight. Building more explainability into AI models remains a technical challenge, but one that is essential for aligning with the principles of transparency and accountability in healthcare.

In the age of ChatGPT and generative AI, misinformation and “AI hallucinations” have emerged as new public health risks. Advanced chatbots can produce remarkably human-like answers to questions — but they do not guarantee factual accuracy. They can hallucinate false information, confidently output incorrect medical advice, nonexistent statistics, or even fake health news. The potential for harm is considerable if the public uses such tools for health information. There is concern that AI chatbots could magnify the health misinformation problem exponentially — for instance, by generating convincing anti-vaccine narratives or spurious cures, which then spread on social media.

In recent years, public health agencies have struggled to combat misinformation (for example, false claims about vaccines or COVID-19 treatments that undermine uptake). The rise of AI-driven content generators and deepfakes only fuels this fire. Misinformation undermines public trust and can lead people to reject proven interventions in favour of dangerous alternatives. Tackling this will require new strategies — such as watermarking AI-generated content, strengthening content moderation, and improving digital health literacy so the public can better discern credible information. On the flip side, public health communicators might also leverage AI to fight misinformation (for example, using AI to detect false rumours early or personalise accurate health messages). Regardless, the advent of easy, AI-generated disinformation is a serious risk factor that the global health community cannot ignore.

Finally, there is the risk of over-reliance and systemic dependency on AI. If health systems come to depend on AI for critical functions without adequate safeguards, any failures in the technology could have severe consequences. For example, an AI model might perform well in normal conditions but fail to generalise during an unexpected scenario. If everyone has come to rely on its output, they may miss the warning signs until too late. Moreover, heavy reliance on automation might erode human skills over time (a phenomenon observed in other industries). In healthcare, this raises concerns about “deskilling” — clinicians might lose practice in specific tasks (like reading x-rays or making complex diagnoses) if those are always handled by AI, leaving them less prepared to step in when needed.

Over-reliance can also dull vigilance: users might stop double-checking results if an algorithm usually works well so that an undetected error could propagate. The key is to maintain a human-in-the-loop approach: AI should support, not replace, human expertise. Mechanisms for human review of AI outputs and fallback plans in case of system outages are essential.

Additionally, performing regular audits and updates of AI models can prevent performance from degrading unnoticed. In summary, while AI can increase efficiency, public health systems must guard against blindly relying on algorithms. A balanced approach that values human judgment and institutional memory, alongside AI’s computational power, will be safest in the long run.

Ethical and Regulatory Frameworks

Addressing the above risks requires robust ethical guidelines and regulatory oversight for AI in health. Globally, there is growing consensus on core ethical principles that should govern AI development and use in public health. The World Health Organization’s landmark 2021 report laid out six guiding principles for ethical AI in health: (1) Protect human autonomy — humans should remain in control of health decisions, with informed consent and respect for privacy; (2) Promote human well-being and safety — AI must be safe, effective, and designed to improve health outcomes; (3) Ensure transparency, explainability and intelligibility — stakeholders should have sufficient information about how AI systems work and decisions should be traceable; (4) Foster responsibility and accountability — developers and users are accountable for AI behaviour, and mechanisms for redress must exist; (5) Ensure inclusiveness and equity — AI should benefit all groups, enhancing fairness and not amplifying disparities; and (6) Promote AI that is responsive and sustainable — meaning AI should be adaptable, monitored, and designed for long-term societal benefit.

These principles, while high-level, provide a value framework to guide everything from design choices (e.g. using diverse training data to ensure equity) to deployment (e.g. always keeping a human in the loop to protect autonomy). Public health organisations are increasingly adopting such ethical frameworks. For instance, the WHO urges that AI deployments be accompanied by community engagement, training for health workers, and continuous evaluation to ensure technologies remain aligned with the public interest. The ethos is straightforward: AI must be people-centred and uphold human rights. Ethics committees or advisory boards can help oversee AI projects, reviewing them for compliance with these principles before they scale up.

On the regulatory front, governments are now moving to establish formal rules for AI in healthcare. The European Union’s AI Act is a pioneering example of comprehensive regulation. Passed in 2024, the EU AI Act takes a risk-based approach, classifying AI systems by risk level and imposing requirements accordingly. Health-related AI is generally deemed “high-risk” under this law, given its potential impact on people’s lives and rights. High-risk AI systems (including most AI used for medical diagnostics, decision support, or resource allocation in health) will face strict obligations. These include rigorous standards for transparency, risk management, and human oversight. For instance, developers of a clinical AI tool must implement a quality management system, ensure their model is trained on appropriate data, and provide documentation detailing the AI’s function and limitations. They must also conduct risk assessments and put in place human oversight measures to prevent automation bias. Notably, the EU AI Act doesn’t just apply to creators of AI — it also holds deployers (such as hospitals or public health agencies) accountable for the safe use of AI.

Health providers must monitor AI system performance, keep logs, and retain ultimate responsibility for decisions (clinicians must have the authority to override AI recommendations if needed). These provisions aim to ensure that human accountability and patient safety remain paramount even as AI becomes embedded in care delivery. Additionally, the Act has a broad reach: any AI system impacting people in Europe must comply, even if developed elsewhere. This could set an effective global benchmark as companies worldwide adjust their practices to meet the EU’s requirements.

Other jurisdictions are also crafting guidelines. The United States, through the FDA, has been evolving its regulatory approach for AI/ML-based medical devices, focusing on premarket evaluation and the idea of “continuously learning” algorithms needing ongoing monitoring. International bodies like the WHO have issued guidance and urged governance innovation, suggesting that governments update regulations to cover AI, establish certification processes, and possibly create registries of approved AI health products. We also see emerging governance models such as algorithmic impact assessments (to evaluate a health AI system’s potential societal impact before deployment) and independent reviewers’ bias audits. In some health systems, procurement of AI now requires meeting ethical checklists or obtaining approval from institutional review boards, similar to new medical interventions.

These steps are part of building a “responsible innovation” culture around AI, encouraging experimentation and advancement, but within guardrails that protect individuals and communities. Multi-stakeholder collaboration is key here — regulators, technologists, health professionals, and patient representatives need to work together to define safe and effective AI in practice and update those definitions as the technology evolves. As one example, the NHS AI Lab in the UK partnered with regulators to create a sandbox for AI developers, guiding them on navigating regulatory pathways and using synthetic data for testing. Such efforts show that with thoughtful governance, innovation and safety can advance hand in hand.

Future Directions and Recommendations

To fully realise AI’s promise in public health while minimising its downsides, several changes and strategic efforts are needed going forward:

• Investing in data and digital infrastructure: Health systems, especially in low- and middle-income countries, need support to build the data foundations for AI. This means digitising health records, improving data quality, and ensuring platform interoperability. Governments and global donors should prioritise funding for health information systems and broadband connectivity as part of public health capacity building. Better data infrastructure not only enables AI — it strengthens health systems overall. Innovative approaches like federated learning (where AI models train on distributed data without moving it) could be scaled to allow resource-constrained regions to benefit from AI insights without breaching privacy. The goal is to create a world where data flows securely and efficiently to wherever it can improve health outcomes.
• Strengthening workforce capacity and AI literacy: As AI becomes a standard tool, public health and healthcare workers must be equipped to use and oversee it. Training programmes are needed to raise AI literacy among the health workforce, including understanding AI’s capabilities and limitations. This may involve updating medical and public health curricula to cover data science basics. Additionally, new specialist roles (such as clinical AI safety officers or epidemiologists with AI expertise) could be developed to bridge the gap between tech and health domains. Frontline staff should be engaged in co-designing AI solutions so that tools are user-friendly and address actual pain points. When health workers understand and trust AI, they can become champions for its adoption and serve as critical watchdogs who notice when something isn’t right. Fostering a culture of continuous human oversight and feedback will ensure that AI remains a servant to health professionals, not a black box dictator.
• Ensuring inclusivity and equity in AI advancement: The global health community must actively work to prevent a digital divide in AI. Much cutting-edge AI development is concentrated in wealthier countries and tech companies. Deliberate efforts are needed to include researchers and perspectives from low- and middle-income countries in AI design so that solutions address diverse needs. This could consist of research funding earmarked for LMIC-led AI projects, technology transfer programs, and south-south collaboration on AI for health. Moreover, data from underrepresented populations should be collected (with consent and protection) to improve algorithms’ relevance in those settings. By democratising AI knowledge and resources, we can avoid a scenario where only certain countries or communities benefit from AI while others are left behind or subject to unchecked harm. Equity considerations should also extend to gender, age, and other demographics — for instance, ensuring women and minority groups are included in AI development teams and that tools serve users of different languages and literacy levels. An inclusive approach will make AI tools fairer and enlarge the talent pool working on creative AI solutions for entrenched public health challenges.
• Fostering collaboration between public health and technology sectors: Effective AI in public health sits at the intersection of epidemiology, medicine, data science, and engineering. No single sector can do it alone. We need stronger partnerships: governments linking with academia and tech firms, NGOs working with startups, and international agencies convening multi-sector consortia for global health AI initiatives. Such collaboration can accelerate innovation and ensure that public health priorities guide technological development (and vice versa, that technologists are aware of on-the-ground needs). For example, a partnership between a national health ministry and AI researchers might focus on building an early warning system for malaria outbreaks, combining epidemiological expertise with cutting-edge modelling. A pharmaceutical company could also collaborate with global health organisations to use AI in vaccine R&D for diseases of poverty. These cross-sector collaborations should be underpinned by fair agreements (e.g. around data sharing or intellectual property) so that all parties benefit and trust is maintained. The complexity of health + AI demands breaking down silos. International forums and networks can play a role here, enabling countries to share best practices and lessons learned (e.g. how one country successfully regulated an AI symptom-checker or how another trained health workers on AI). Since pathogens do not respect borders, a collaborative global approach to AI-enhanced public health security is in everyone’s interest.
• Adaptive governance and continuous evaluation: As AI tools roll out, it is critical to monitor their real-world impact and be ready to adjust course. Public health authorities should implement mechanisms to continuously evaluate AI interventions — collecting data on their accuracy, outcomes, and any unintended effects. Are the predictions helping improve disease control? Is a triage algorithm safely directing patients to the right level of care? This requires establishing key performance indicators and perhaps creating independent evaluation units. When problems are identified (such as an AI starting to drift in accuracy due to changes in data), there should be processes to update or pull back the tool until fixes are in place. Regulation must also remain adaptive; rigid rules could stifle innovation or become outdated as technology advances. One idea is regulatory sandboxes where new AI solutions can be tested under supervision, allowing regulators to learn and guidelines to evolve. Governance models should be proactive yet flexible, emphasising learning and iteration. Importantly, communities and civil society should have a voice in evaluating AI in public health — their feedback on whether these tools are culturally acceptable, understandable, and improving services is invaluable. Responsible AI is not a one-time certification but an ongoing commitment to quality and ethics throughout the technology’s lifecycle.

Looking ahead, it is clear that AI will play an expanding role in public health — whether in combating the next pandemic, extending healthcare to remote villages via smart apps, or analysing big data to pinpoint disease drivers. The revolution is already underway, but its trajectory depends on our current choices. With enlightened leadership, adequate safeguards, and inclusive collaboration, AI could usher in significant public health gains — from more efficient health systems to healthier communities worldwide. However, if we ignore the risks — allowing unchecked use, widening inequities, or losing the human touch in care — the potential benefits could unravel, and public trust could be irrevocably lost. The coming years are thus pivotal. Armed with decades of hard-won experience, public health professionals have a key role in steering this journey. By insisting on evidence, equity, transparency, and community engagement, they can ensure that the AI revolution in health truly becomes a boon and not a threat. The opportunity is immense, but so is the responsibility to guide AI’s integration into public health thoughtfully and ethically.