Leveraging artificial intelligence for One Health: opportunities and challenges in tackling antimicrobial resistance - scoping review

Abstract

Background

Antimicrobial resistance (AMR) is a global health challenge driven by the misuse of antimicrobials across humans, animals, and the environment, necessitating integrated One Health solutions.

Objective

This scoping review aims to synthesise evidence on the opportunities and challenges of leveraging artificial intelligence (AI) to tackle AMR within the One Health framework.

Methods

This review adhered to the PRISMA-ScR guideline. A comprehensive literature search was conducted in PubMed, Embase, Scopus, Web of Science, along with citation searching and Google Scholar.

Results

A total of 543 studies were identified from these databases. After removing duplicates, 343 studies remained for screening. Following the title and abstract screening, 273 publications were selected for full-text review, and 43 studies were included in the final analysis. Studies written in English that explored the application of AI tools and techniques for AMR in any One Health domain were included. The review found that AI is widely applied to combat AMR across different sectors (human, animal, and environmental health), with key opportunities including the rapid identification of resistant pathogens, AI-powered surveillance and early warning, integration of diverse datasets, and support for drug discovery and antibiotic stewardship. However, significant challenges remain, such as data standardisation issues, limited model transparency, infrastructure and resource gaps, ethical and privacy concerns, and difficulties in real-world implementation and validation.

Conclusion

Overall, while AI has great potential to improve AMR management, fully realising its benefits will require investment in explainable AI, better data infrastructure, stronger cross-sector collaboration, and clear regulatory frameworks to ensure ethical and effective use within the One Health approach.

Supplementary Information

The online version contains supplementary material available at 10.1186/s42522-025-00170-8.

Introduction

Antimicrobial resistance (AMR) is a critical global health challenge, projected to cause 10 million deaths annually by 2050 if left unaddressed [1]. Research indicates that the misuse and overuse of antimicrobials in clinical, agricultural, and environmental settings have accelerated the evolution and dissemination of resistant pathogens, threatening the efficacy of life-saving treatments [2]. Additionally, releasing antimicrobial-resistant bacteria into the environment from sources such as pharmaceutical factories, hospitals, agricultural farms, and wastewater contributes to the formation of resistance gene reservoirs. These genes can then be transmitted to human pathogens via water, soil, and food [3]. This illustrates the need for integration among multiple sectors to combat the issue of AMR. The One Health framework integrates human, animal, and environmental health to address AMR’s complex and interconnected drivers across ecosystems [4]. This approach recognises that the health of people, animals, and the environment is deeply interdependent and that AMR cannot be effectively managed by focusing on a single sector alone [4]. The Quadripartite Collaboration (Food & Agriculture Organisation (FAO), World Health Organisation (WHO), World Organisation for Animal Health (WOAH), United Nations Environment Program (UNEP)) has developed a One Health-driven joint plan of action to integrate systems and capacity for AMR surveillance, governance framework, and cross-sectional interventions worldwide [5, 6].

Timely integrated data on AMR and antimicrobial use across human, animal, and environmental sectors is critical to curbing AMR’s devastating impact [1]. The One Health approach requires combining and analysing varied and heterogeneous data streams and types [4]. The collaboration among multiple sectors necessitates handling large volumes of data and robust tools for extracting valuable insights from complex datasets. To this end, artificial intelligence (AI) approaches provide new avenues for cross-sectoral evaluation of existing and future health threats [7]. The rapid advancement of AI offers transformative potential for healthcare by analysing large-scale datasets, enabling early detection of resistance markers, optimising antimicrobial use through predictive analytics, and accelerating drug discovery by identifying novel antibacterial agents [8]. For instance, AI systems can integrate genomic data with environmental surveillance to predict resistance hotspots and guide targeted interventions [4]. Moreover, initiatives like the Outbreak Consortium demonstrate the potential of AI-driven platforms to map risks using diverse datasets from various sectoral records, thereby fostering more effective antimicrobial stewardship [5]. Given the complexity of AMR and the promise of AI-enhanced solutions, it is essential to understand how these technologies are currently being utilised within One Health frameworks. A review conducted by Howard et al. [9] highlights that AMR is influenced by One Health factors, such as agricultural antimicrobial use and wastewater management, which often operate independently with distinct governance, dataflows, and priorities, and require collaborative One Health data science networks to align and integrate these data streams [10]. Existing surveillance systems often operate in silos, with limited capacity to combine and analyse heterogeneous data streams from multiple sectors, hindering timely detection of resistance trends and the development of comprehensive, evidence-based strategies [11]. Most existing syntheses focus on single sectors, such as human clinical data or bacterial genomics, and do not systematically address how AI can be leveraged to integrate and analyse cross-sectoral data from human, animal, and environmental health [12, 13]. Additionally, while AI models have shown the ability to identify resistance patterns, predict treatment outcomes, and accelerate antibiotic discovery, there is a lack of reviews that critically assess the real-world implementation challenges, data interoperability, and the potential for AI-driven decision support in multi-sector AMR surveillance [9]. There is an urgent need for advanced analytical tools, such as AI, that can integrate and interpret large-scale, multisectoral datasets to enable early detection of resistance hotspots, optimise antimicrobial use, and inform targeted interventions within the One Health paradigm. To date, there appears to be no comprehensive synthesis that has mapped AI’s potential applications, challenges, and gaps within the One Health framework. Therefore, this scoping review aims to synthesise the currently available evidence on the opportunities and challenges of integrating AI to combat AMR within the One Health approach to inform future research, policy, and resource allocation.

Research questions

1. What are the current applications of AI in AMR surveillance across human, animal, and environmental health sectors?

2. What key opportunities does it present for enhancing AMR monitoring within the One Health framework?

3. What challenges hinder the integration of AI in AMR monitoring with the One Health framework?

Methods

Protocol and registration

This scoping review was conducted following the Preferred Reporting Items for Systematic Reviews and Meta-analysis Extension for Scoping Reviews (PRISMA-ScR) guidelines [14] and the Joanna Briggs Institute (JBI) guidance [15] to map the nature and characteristics of the existing studies focused on the application and opportunities of AI to combat AMR within a One Health approach. The protocol of this review was registered in the Open Science Framework (OSF) platform https://osf.io/rm4w5/.

Search strategy

A comprehensive literature review was conducted through PubMed, Embase, Scopus, and Web of Science, along with citation searching and Google Scholar. These databases are widely recognised and complementary in their search for biomedical and clinical-related studies, helping to minimise publication bias and increase the likelihood of capturing all relevant studies for our review. The search strategies were drafted by the first author and further refined through team discussion. The search was conducted between February 10, 2025, and April 15, 2025. The search timeline from 10 February 2025 to 15 aligned with the timeline of our research activities, including protocol finalisation and database access. We used different search terms and keywords, “artificial intelligence”, “machine learning”, “deep learning”, “global health”, “antimicrobial resistance” or “antibiotic resistance”, “drug resistance”, “One Health”, “animal health”, “human health”, “environmental health”, and “human-animal-environment”. Additionally, Boolean operators (AND, OR) were employed to include synonyms or closely related terms within the same concept and combine different concepts to narrow the results to studies that address all key aspects of the research question. In Google Scholar, we employed a variety of search techniques to ensure comprehensive coverage of relevant literature. These included targeted keyword searches using terms directly related to our research question. After searching the database, all identified citations were imported into EndNote and duplicates were removed. To identify grey literature, we screened the bibliographies of selected studies and reviewed relevant organisational and government websites. The completed search strategies for each database are available in Supplement File 1.

Eligibility criteria

The inclusion criteria for the review were original research published in peer-reviewed journals, conference papers, book chapters, review articles, and grey literature written in English that explore the application of AI tools and techniques for AMR or related antimicrobial use and misuse across different sectors (human, animals, and the environment) within One Health domains, not necessarily as their primary focus. To ensure a comprehensive mapping of the available evidence, our inclusion criteria were designed to encompass a broad range of literature types. This approach aligns with scoping review best practices, which aim to capture all relevant sources of evidence and provide a holistic overview of the field. Articles that do not explore or investigate the use of AI for AMR in at least one of the One Health disciplines, and those published in languages other than English, were excluded.

Study selection and data charting process

The identified studies were imported into EndNote 20.3. EndNote was used to remove duplicates because it is a reliable and widely used reference management tool that efficiently identifies and eliminates duplicate records. After the removal of duplicates, four authors independently conducted the study selection process in two steps. In the first step, the authors evaluated only the titles and abstracts for relevance. In the second step, the full text of all publications that met the inclusion criteria during the first screening was retrieved for further screening. A data charting form was developed by the first author (GK) and refined by the other three authors (SI, JH, and SC). Data from each included study were extracted into a data extraction form that included the author, year of publication, study country /region, objectives, study type, One Health domain addressed, type of AI, key findings/outcomes, challenges reported, and suggested recommendations. We resolved any disagreements on study selection and data extraction by discussion.

Synthesis of results

After selecting the eligible studies, each article was assessed for its key findings, reported challenges, and recommendations, based on the research objectives and questions. The extracted information was recorded in an Excel sheet. Using an inductive approach, different codes were then generated from the data. These codes were grouped into initial candidate themes that captured meaningful patterns. Similar themes were later merged and refined to produce the final themes. In total, eight themes related to the opportunities and applications of AI for AMR were identified, along with six themes addressing challenges to its implementation. Each theme was described narratively.

Results

Search results

A total of 543 studies were initially identified. Of these, 473 articles were retrieved from Web of Science, Scopus, PubMed, and Embase, while 70 studies were identified through citation searching and Google Scholar. After removing duplicates, 343 studies were screened based on their titles and abstracts, and 278 studies were retrieved for full-text review. Following the assessment of the inclusion criteria, 43 studies were included in the final review (Fig. 1).

Fig. 1.

Flow diagram of study selection

General characteristics of included studies

A total of 43 studies were included in the review. These studies explored AI applications such as prediction models, clinical decision support, surveillance, drug discovery, and analysis. Most studies took a global perspective, while eight studies were conducted in specific countries [16–24] (Table 1). The majority of publications on AI integration in combating AMR focus on human health applications. Conversely, only a few studies report on AI use in animal health and environmental contexts [18, 23–29]. The common AI methods described in these studies include Naïve Bayes, Decision Trees, Random Forests, Support Vector Machines, ML, and Artificial Neural Networks, applied across human, animal, and environmental health domains. Collectively, all studies highlight the opportunities, challenges, and potential impacts of AI in addressing AMR across various sectors and offer recommendations to harness AI for combating AMR through human, animal, and environmental health, as shown in Table 1.

Table 1.

Characteristics of included studies

Author/year	Study country/region	Study type	Study objectives	Types of AI	One Health area	Key findings/outcomes	Challenges reported by the included Studies	Suggested recommendation by included studies
Abavisani et al. [16]	Global	Review	To comprehensively examine the opportunities and challenges presented by chatbots in bacterial disease management and ABR mitigation.	Chatbots	Not specified	⎫ Enhanced disease management and patient engagement: Chatbots can support bacterial disease management by aiding in symptom assessment, medication adherence, mental health support, and fostering patient engagement. ⎫ Chatbots can provide accessible educational support to improve awareness about antibiotic use, thereby contributing to antibiotic resistance (ABR) mitigation.	⎫ Ethical, legal, and privacy concerns: The integration of chatbots must address challenges related to user data privacy, consent, and compliance with ethical and legal standards. ⎫ The effectiveness of chatbots can be hindered by technical limitations, such as inaccuracies in information or misinterpretation of user input, leading to potential patient harm or mistrust.	⎫ Researchers, healthcare professionals, and policymakers must work together to guide the safe and effective integration of chatbots into healthcare systems. ⎫ Increased funding and regulatory frameworks are needed to improve chatbot technology, ensuring it remains ethical, secure, reliable, and tailored to healthcare needs.
Acharjee et al. [17]	Global	Review	Assessing the accuracy and dependability of ML Models that utilise genomic data to predict emerging resistance trends.	ML	Not specified	⎫ Enhanced predictive power and diagnostics: AI improve early prediction of resistant pathogens and enables rapid, accurate diagnostics, which help guide timely and effective interventions. ⎫ Accelerated drug discovery and optimisation: AI facilitates the discovery of new antibiotic compounds and optimises existing treatments, enabling faster and more efficient development of next-generation antimicrobial agents.	⎫ There is a gap between AI-driven innovations and their practical application in clinical settings, due to a lack of interdisciplinary collaboration and integration into healthcare systems.	⎫ Invest in Sustained AI Research: Continued research and development are essential to advance AI capabilities for AMR, ensuring tools remain adaptive and effective. ⎫ Foster Interdisciplinary Collaboration, it is vital to effectively implement AI innovations in real-world healthcare settings.
Akinsulie et al. [18]	Global	Review	To highlight and discuss the potential impact of various aspects of AI in veterinary clinical practice and biomedical research, proposing this technology as a key tool for addressing pressing global health challenges across various domains.	ML	Animal health	⎫ AI enables rapid identification of resistant strains, accelerates antibiotic discovery and supports predictive analytics for outbreak detection and targeted treatment. ⎫ Application across sectors (One Health Approach): AI can be used in both human and veterinary health to detect antibiotic residues in food, track resistance genes, and guide antibiotic use policies, supporting a comprehensive AMR control strategy.	⎫ Concerns around informed consent, data privacy, and ethical use of patient and genetic data limit AI’s full integration into healthcare systems. ⎫ Heterogeneity, bias in electronic health records, limited examples for training algorithms, and the “black-box” nature of many AI models reduce reliability, interpretability, and generalizability.	⎫ Emphasis should be placed on explainable AI (XAI), real-time data analysis, and cross-disciplinary collaboration to improve accuracy and trust. ⎫ Promote the use of structured databases to improve training data quality, foster algorithm development, and facilitate integration of AI in both human and animal health diagnostics.
Ali et al. [19]	Global	Review	To understand the challenges and opportunities of harnessing AI to combat AMR	ML	Not specified	⎫ Availability of large antimicrobial-related datasets and powerful computing infrastructure enables DL and ML applications for AMR research. ⎫ AI research in antibiotics is opening new pathways for solving challenges in diagnostics, resistance prediction, and environmental monitoring.	⎫ Limited AI expertise. ⎫ Most AI models rely on imbalanced or geographically biased data, which limits generalizability and clinical relevance.	⎫ Develop interactive AI models to capture the complexity of features influencing AMR development and resistance patterns. ⎫ Ensure balanced, diverse, and standardised datasets to improve model reliability, accuracy, and broader applicability across different regions and healthcare systems.
Anahtar et al. [20]	Global	Revie w	To review the current state of research at the intersection of ML and AMR	ML	Not specified	⎫ Enhanced AMR management through ML: advanced ML algorithms offer new pathways to improve diagnostics, understand resistance mechanisms, accelerate antibiotic development, and support personalised treatment decisions. ⎫ Growing institutional and computational support: the continued rise in data generation, computational power, algorithm sophistication, and institutional interest (from governments and public-private partnerships) suggests strong future potential for ML-driven AMR solutions.	⎫ Limited high-quality training data: a lack of robust, high-quality datasets hampers the development and accuracy of ML models in the AMR domain. ⎫ Insufficient interdisciplinary Collaboration.	Build interdisciplinary research teams: Encourage collaboration among experts to foster innovation and holistic solutions to AMR.
Arnold et al. [3]	Global	Review	To explore the integration of AI across three critical domains essential for curbing the escalation of AMR: clinical diagnostics, AMR surveillance and antibiotic discovery.	Not specified	Not specified	⎫ Growing high-quality data resources are creating robust, annotated, and standardised datasets that enable AI applications in AMR prediction and antibiotic discovery. ⎫ Advances in Chemical Screening Data: the increasing availability of chemical screening data supports AI-driven identification of potential antibiotics, particularly for early-stage drug discovery efforts.	⎫ Lack of diverse and standardised data hinders the development of generalizable, clinically applicable AI models. ⎫ Insufficient Experimental Validation: many AI models, especially in surveillance and drug discovery, lack real-world validation, relying heavily on test metrics rather than confirming predictions through wet lab or clinical testing.	⎫ AI predictions in AMR and antibiotic discovery must be followed up with rigorous in vivo and clinical testing to confirm safety, efficacy, and clinical utility. ⎫ Foster cross-disciplinary collaboration: successful deployment of AI in AMR requires strong partnerships among data scientists, microbiologists, clinicians, and policymakers to ensure ethical, interpretable, and effective applications.
Ayesiga et al. [21]	Sub-Saharan Africa	Review	To review explores what is known about the integration and applicability of AI-enhanced bio surveillance methodologies in sub-Saharan Africa,	Naïve Bayes, Decision trees, Random Forests, Support Vector Machines and Artificial Neural Networks	Not specified	⎫ Enhanced AMR Biosurveillance: AI technologies offer transformative potential in detecting, tracking, and predicting resistant strains. ⎫ Improved collaboration and understanding: AI can facilitate a holistic understanding of AMR dynamics in different sectors (human, animals and the environment) through secure, interoperable data-sharing platforms and integration of diverse data sources.	⎫ Data Limitations: lacks large, high-quality datasets for training AI models and often relies on data from high-income countries, reducing the relevance and robustness of AI applications in the region. ⎫ Ethical and infrastructural challenges.	⎫ Context-specific AI development: develop AI models tailored to local epidemiological and infrastructural contexts to increase effectiveness and relevance. ⎫ Establish and implement regional ethical guidelines and secure data-sharing platforms to ensure responsible AI deployment and foster cross-country collaboration.
Baker et al. [22]	China	Original article	By using a data mining approach based on ML, to identify potentially mobile antibiotic resistance genes (ARGs) shared between chickens and environments across all farms.	ML and metagenomic analysis	Animal and environmental	⎫ There is an intricate network of correlations between environments, microbial communities and AMR, suggesting multiple routes to improving AMR surveillance in livestock production. ⎫ Metagenomic sequencing has the potential to broaden our knowledge of the factors driving resistance and improve AMR surveillance.	⎫ Methodologies need to be standardised, and data gaps filled. ⎫ Only a few laboratories and countries at present have both resources and expertise to use metagenomic sequencing for AMR surveillance.	With further development, metagenomic and ML approaches could be deployed to provide fast and reliable prediction of AMR outbreaks, emerging pathogens and transmission routes.
Branda and Scarpa [23]	Global	Review	To understand the implications of AI in addressing AMR, and its challenges.	ML	Not specified	⎫ AI accelerates genomic analysis to detect resistance markers early and facilitates the discovery of new antibacterial agents through predictive modelling and computational simulations. ⎫ AI enables personalised antibiotic regimens, optimises antibiotic dosing (e.g., vancomycin, amikacin, colistin), and supports real-time clinical decision-making using AI-powered systems.	⎫ Data heterogeneity and quality issues: a major challenge is the lack of standardised, annotated, and high-quality data across sources, which compromises model accuracy and generalizability. ⎫ Lack of interpretability and transparency: many AI models act as “black boxes,” creating mistrust and resistance among healthcare providers due to their opaque decision-making processes.	Establish centralised, high-quality data repositories with standardised formats to enhance model training and performance across clinical settings. Use explainability tools to ensure algorithmic fairness and engage in independent validation and regulatory oversight to build trust and support ethical deployment of AI in healthcare.
Cesaro et al. [24]	Global	Review	To explore AI applications in diagnostics, therapy, and drug discovery, emphasising both strengths and areas needing improvement.	ML, DL	Not specified	⎫ AI can analyse vast clinical and experimental datasets to improve disease diagnosis, predict treatment outcomes, and accelerate the discovery of novel antimicrobial compounds. ⎫ Speeding up antibiotic discovery, offering tailored treatment regimens and reinforcing antimicrobial stewardship.	⎫ Data bias and limited diversity: many AI models are trained on imbalanced or biased datasets, limiting generalizability and potentially perpetuating inequities across patient populations. ⎫ Ethical and trust issues: Concerns around data privacy and informed consent hinder trust and transparency in AI-driven decisions.	⎫ Ensure datasets are representative, well-annotated, and curated with attention to ethical standards to improve AI model reliability and fairness. ⎫ Promote partnerships among healthcare professionals, developers, and regulators to build transparent, interpretable, and ethically sound AI systems.
Chen et al. [25]	Global	Review	To understand what and how AMR research was being performed now, to identify knowledge gaps and guide future integrated One Health research activity	Natural Language Processing	Not specified	AI approach rapidly, accurately, and efficiently categorised information from diverse sources, systematically characterising AMR research around the world over time, across income statuses, and One Health sectors.	⎫ Poor communication between sectors. ⎫ Lack of cross-sectoral research, especially linking plant health and others.	⎫ Performing culture-based and genomic AMR analysis in tandem in all sectors is crucial for data integration and holistic One Health solutions. Increased investment in capacity development.
Condorelli et al. [26]	Italy	Original article	To predict AMR by combining and applying ML methods to bacterial genomic data, specifically for Klebsiella pneumoniae, to support clinicians in selecting appropriate therapy	ML	Human health	⎫ ML can predict AMR early and personalise antibiotic treatments, improving efficacy and reducing misuse. ⎫ Accelerated drug discovery and research efficiency: ML can analyse large chemical databases to identify potential new antibiotics and streamline research efforts, saving time and resources.	Lack of integration of contextual factors like geographical location, clinical setting, and patient history can reduce model accuracy and generalizability.	Further research is needed to understand complex biological interactions and enhance the interpretability and accuracy of ML models in AMR studies
de la Lastra et al. [27]	Global	Review	To explore the role of AI/ML in AMR management, with a focus on identifying pathogens, understanding resistance patterns, predicting treatment outcomes, and discovering new antibiotic agents	ML	Not specified	⎫ Enhanced prediction and personalisation: AI/ML can predict AMR trends, optimise drug administration, and customise treatments using integrated genomic, phenotypic, clinical, and epidemiological data. ⎫ Accelerated discovery and decision support: These technologies support rapid identification of resistance mechanisms, assist in diagnosing infections, and facilitate the development of new therapies with minimal human intervention.	⎫ Data Limitations: limited access to comprehensive and high-quality datasets (e.g., biased, incomplete, or non-standardised) undermines prediction accuracy and model performance. ⎫ Transparency and Ethical Concerns: many AI/ML models lack interpretability (black-box issue), and concerns about data privacy, regulatory compliance, and clinical accountability limit adoption in healthcare.	⎫ Develop standardised, diverse, and interoperable datasets to enhance the reliability and accuracy of AI/ML models in AMR research. ⎫ Foster Interdisciplinary Collaboration: encourage collaboration between computer scientists, healthcare professionals, biologists, and ethicists to ensure the safe, ethical, and effective implementation of AI in clinical and research settings.
Duggirala et al. [28]	USA	Review	To explore the Centre for Veterinary Medicine’s strategic approach to harnessing AI technologies to enhance human and animal health.	ML	Animal and human health	⎫ Regulatory Innovation: AI offer potential for more adaptive, efficient, and innovative regulatory decision-making processes. ⎫ Enhanced surveillance and research tools: These technologies support improved detection of safety signals, more precise case processing, and deeper insights into genomic and AMR data.	⎫ Lack of integration. ⎫ Data quality and model validation issues.	Strengthen Collaboration: Encourage ongoing partnerships between researchers, regulatory bodies, and industry stakeholders to drive effective AI integration.
Elalouf et al. [29]	Global	Review	To highlight the application of AI, particularly DL and ML in managing AMR.	ML, DL	Not specified	⎫ AI have significantly advanced the prediction of drug resistance patterns and the identification of novel antibiotics. ⎫ AI’s ability to process large, complex datasets allows for deeper insights into AMR trends and tailored interventions across diverse settings. ⎫ Integration of AI in surveillance and decision-making can improve disease management through real-time data analysis and personalised healthcare strategies.	⎫ Data scarcity and heterogeneity limit the accuracy and generalizability of AI models in predicting resistance patterns. ⎫ Ethical and privacy concerns, including a lack of model transparency and risk of worsening healthcare disparities, pose challenges to wider AI adoption.	⎫ Adopt a multi-sectoral, One Health approach involving healthcare, environment, and community sectors, with strong surveillance and data sharing mechanisms. ⎫ Develop transparent and interpretable AI models while addressing data quality, ethical issues, and privacy concerns to maximise AI’s effectiveness in AMR control.
Elyan et al. [30]	Global	Review	Evaluation of utilising ML to tackle AMR and discussion of challenges and limitations that are considered barriers to scaling up the use of ML to address AMR	ML, DL	Not specified	Enhanced cross-sectoral collaboration by providing a data-driven framework to improve antibiotic prescribing practices in LMICs, where management of AMR remains most challenging.	⎫ Access to high-quality data remains a major challenge for ML and DL in AMR research, as most models rely on limited, proprietary, or geographically restricted datasets. ⎫ Issues like class imbalance and concept drift, where AMR data evolve further, complicate model development and reliability.	A future direction should be towards more coordinated efforts between the different stakeholders and ongoing recording and monitoring of AMR-related data.
Fanelli et al. [31]	Global	Review	To show the potential role of AI in fighting AMR in paediatric patients.	ML	Human health	AI is contributing to a reduction in the development time of new antimicrobial agents, greater diagnostic and therapeutic appropriateness, and, simultaneously, a reduction in costs.	The lack, especially in the pediatric field, of randomised clinical trials that demonstrate the reliability and/or improved effectiveness of AI systems compared to traditional systems in diagnosing infectious diseases or suggesting appropriate therapies creates a certain mistrust on the part of physicians toward the use of systems based on AI.	⎫ Further studies are needed on ethical, regulatory, or practical considerations required for AI widespread use. ⎫ The global health community will need to work quickly to establish guidelines for development, testing, and use, as well as to develop a user-driven research agenda to facilitate its equitable and ethical use.
Ferrari et al. [32]	Italy	Original article	To predict bloodstream infection and AMR in patients using a ML approach.	ML	Human health	Advanced ML models effectively predicted AMR at ICU admission using existing patient data, outperforming older methods, and highlighting potential for future improvements despite stagnant stewardship practices.	⎫ Limited dataset and data imbalance ⎫ Lack of external validation and interpretability; the absence of external validation and the challenge of making ML predictions interpretable for clinicians hinder clinical adoption and scalability.	⎫ Expand the Dataset ⎫ Increase the dataset’s size and diversity by involving multiple institutions ⎫ Collaborate with Clinicians for Integration and Interpretability
Foroughi et al. [33]	Global	Review	To investigate the application of ML-based methods to solve AR-induced problems in water and wastewater.	Shallow learning, DL,	Environment health	⎫ Applying underused MLM techniques (e.g., clustering, association, and DL) can offer new insights into AR-related challenges in water and wastewater systems. ⎫ The growing complexity of AR problems and non-linear phenomena in environmental media makes ML a promising tool for improving the prediction and understanding of AR dynamics.	Limited use of advanced ML techniques, particularly DL and unsupervised methods, restricts the analytical depth and potential of current research.	⎫ Address existing knowledge gaps by expanding research into clustering, association analysis, and DL approaches for more comprehensive AR investigations. ⎫ Promote broader adoption of MLMS as robust analytical tools to explore various dimensions of AR in water and wastewater, aligning with One Health strategies.
Giacobbe et al. [34]	Global	Review	To review and discuss the use of ML algorithms for antimicrobial stewardship interventions, highlighting opportunities and challenges.	ML	Not specified	⎫ Improved predictive accuracy: ML models, particularly complex ones, can outperform interpretable models in predicting AMR and recommending targeted therapies, enhancing antimicrobial stewardship efforts. ⎫ Support for clinical decision-making: when effectively implemented, ML-powered Clinical Decision Support Systems (ML-CDSSs) can assist clinicians in making evidence-based, data-driven treatment decisions.	⎫ Lack of explainability: the “black box” nature of many accurate ML models reduces clinician trust and poses risks of undetected harmful recommendations due to limited model transparency. ⎫ Data quality and transparency Issues.	⎫ Prioritise interpretability in model design ⎫ Ensure data transparency and legal compliance: model development must consider data quality from the start and adhere strictly to privacy regulations and AI governance laws in healthcare.
Hatim et al. [35]	Global	Review	To assess fundamental concepts of AI, the resources now available for AI, its uses and scope, as well as its benefits and limits	ML, DL	Not specified	⎫ AI can process vast and diverse datasets (e.g., genetic sequences, medical records, epidemiological data) to detect new AMR strains early and monitor their spread. ⎫ AI-driven diagnostic tools support real-time decision-making, promoting public health protection and more responsible antibiotic use.	Limited integration and standardisation of diverse data sources may hinder AI effectiveness in AMR detection.	⎫ Enhance the development and deployment of AI tools to support routine AMR diagnosis and surveillance across various healthcare systems. ⎫ Encourage responsible and ethical use of AI technologies, ensuring they are accessible, interpretable, and used to support evidence-based antimicrobial stewardship.
Ito et al. [36]	Japan, USA, UK, Australia, Sweden, Brazil, Taiwan, Russia	Original article	To explore the possibility that the ethical dilemmas inherent in the context of AMR may hinder the adoption of diagnostic AI (AI).	Artificial Neural Networks	Human health	AI, particularly ML and DL, offers powerful tools for addressing the complexities of AMR. These technologies can process vast amounts of data, identify patterns, and make predictions that would be impossible for humans to achieve manually.	⎫ Implementation Challenges: Regulatory, organisational, and data processing hurdles complicate the integration of AI systems into healthcare environments.	⎫ Ensure that AI implementation is supported by a comprehensive regulatory, organisational, and technical infrastructure to optimise its use in AMR control. ⎫ Promote continued research and global collaboration: advance the field through international cooperation, responsible innovation, and the integration of emerging technologies to enhance AI’s impact on AMR.
Kolluru et al. [37]	Global	Review	To highlight some of the remarkable opportunities that ML offers when applied to research related to AMR.	ML	Human health	⎫ ML (ML) enables accurate prediction of AMR and accelerates drug discovery, improving surveillance and clinical decision-making ⎫ ML models can integrate genomic, phenotypic, and other data sources to support antimicrobial stewardship and inform public health strategies.	Data-Related Challenges: Issues such as limited data availability, poor quality, and lack of standardisation hinder the effectiveness and scalability of ML applications. Ethical and Regulatory Concerns.	⎫ Promote interdisciplinary collaboration: successful integration of ML into healthcare requires partnerships among clinicians, researchers, and computational experts. ⎫ Ensure responsible use and data governance: Addressing data quality, availability, standardisation, and ethical considerations is essential for trustworthy and sustainable ML applications in AMR research.
Lau et al. [38]	Global	Review	To provide an updated overview of antimicrobial design workflow using the latest machine-learning antimicrobial discovery algorithms in the last 5 years	ML	Not specified	AI has significantly improved the procedures of AMR identification, antibiotics development, and discovery by significantly reducing resources, time and effort compared to that of traditional methods.	Despite the significance of AMR and climate change, few AI-focused studies have been published recently, indicating a research gap	Strengthen interdisciplinary collaboration: Combine expertise from various fields and exploit complementary data sources to overcome biases and maximise the impact of AI in public health.
Liu et al. [39]	Global	Review	To review the involvement of AI in antibacterial drug development and utilisation.	DL	Not specified	⎫ AI accelerates the discovery of novel antimicrobial agents, drug repurposing, and resistance mechanism prediction, significantly reducing time and cost. ⎫ Future integration in drug development stages: AI is expected to play a larger role in molecular design, dosage prediction, toxicity reduction, and modelling interactions, improving the clinical success rate of antibiotics.	⎫ Limited interpretability of AI models: Many AI systems, especially DL models, function as “black boxes,” making their decision-making processes hard to understand and trust. ⎫ AI models may rely on superficial patterns (shortcut learning), leading to failure under complex or novel conditions, especially in late-stage drug design.	Increase expert oversight and interpretability: implement expert supervision and methods to interpret model decisions to enhance reliability and trust in AI systems.
Lv et al. [40]	Global	Review	To briefly introduce how to employ AI technology against AMR by using the predictive AMR model, as well as future research directions.	Naive Bayes, Decision Trees, Random Forests, Support Vector Machines	Human, animal and environment	⎫ AI-based methods have significantly improved the speed and precision of identifying AMR (AMR), thereby supporting individualised treatment strategies. ⎫ The development of comprehensive and regularly updated AMR databases ⎫ The integration of multiple biomarkers and factors to enhance the accuracy of predicting treatment effectiveness and guiding clinical decision-making.	⎫ Lack of standardised, frequently updated, and shared AMR datasets. ⎫ Existing AI models often rely on binary classification and struggle with high-dimensional data, limiting predictive accuracy and generalizability across different datasets.	⎫ Apply advanced AI techniques like transfer learning and few-shot learning to improve model generalizability and support more robust AMR prediction in diverse scenarios. ⎫ Automatic annotation of unlabeled data using unsupervised learning is a direction for the future study.
Massé et al. [41]	Canada	Article	To describe risk factors associated with AMR on dairy farms.	Conventional and unsupervised AI	Animal health	By using unsupervised AI analysis, demonstrate the significant risk factors for the presence of AMR on the farms.	Complexity of analysing risk factors associated with AMR.	Further prospective studies with particular emphasis on biosecurity measures are needed.
Masud et al. [42]	Global	Review	To highlight some of AI’s successful uses and potential to combat AMR.	ML, DL	Not specified	⎫ AI enables rapid and accurate AMR pathogen detection by analysing large-scale, multidimensional data (e.g., genomic, clinical, epidemiological). ⎫ Real-time surveillance & decision support: AI-powered tools improve antimicrobial susceptibility testing, treatment selection, and early detection of emerging AMR strains, supporting timely interventions.	⎫ Lack of high-quality, labelled data and the difficulty of forecasting high-dimensional traits limit AI accuracy and generalizability. ⎫ High implementation costs, lack of standardised data, limited digital infrastructure, and shortage of skilled personnel hinder AI integration in clinical settings.	⎫ Promote global collaboration on data sharing, and establish regulatory frameworks to ensure ethical, secure, and unbiased AI use. ⎫ Capacity building: train healthcare workers in AI tools and significantly increase funding for AMR-related AI research through government and industry support.
Mohammed et al. [2]	Global	Review	To explore the burgeoning role of AIin revolutionising AMR strategies.	ML, DL, Natural Language Processing Computer vision	Not specified	AI-driven tools and methodologies can enhance the detection, treatment, and prevention of AMR through various means, including antibiotic discovery, diagnostic and susceptibility testing, and epidemiology and infection control.	⎫ Technical, ethical, data-related, and systemic issues ⎫ Lack of standardisation across data sources and the underrepresentation of certain populations.	⎫ Developing clear ethical guidelines, robust data governance frameworks, and interoperable healthcare technologies, ⎫ Moreover, fostering interdisciplinary collaboration and engaging with the public to build trust are crucial steps toward leveraging AI’s full potential in the fight against AMR.
Mohseni and Ghorbani [43]	Global	Review	To examine various applications of AI in microbiology, including activities such as predicting drug targets and vaccine candidates, identifying microorganisms responsible for infectious diseases, classifying drug resistance to antimicrobial drugs, predicting disease outbreaks.	Naive Bayes, Decision Trees, Random Forests, Support Vector Machines	Not specified	⎫ AI enhances predictive modelling, drug discovery, vaccine development, pathogen detection, and personalised medicine in microbiology. ⎫ It supports microbiome research, environmental monitoring, rapid diagnostics, and synthetic biology, contributing to improved public health outcomes.	⎫ Heterogeneity and poor quality of microbiological data hinder effective AI analysis and model training. ⎫ Lack of transparency: many AI models are “black boxes,” making their predictions hard to interpret, which limits trust and usability in sensitive biomedical contexts.	⎫ Interdisciplinary collaboration: encourage cooperation among researchers, practitioners, and policymakers to promote responsible AI use and best practice sharing.
Monaco et al. [44]	Italy	Original article	To employ the explainable AI paradigm to identify the factors that most affect the onset of AMR in diversified territorial contexts, which can vary widely from each other in terms of climatic, economic and social conditions through One Health frameworks.	ML	Environmental health	⎫ Scalability and Generalization: The decision support tool developed can be generalized to other pathogens and extended to include additional environmental and social factors, broadening its applicability. ⎫ Potential for Enhanced Monitoring: Incorporating data collected at different times could support the development of dynamic monitoring procedures for AMR trends.	⎫ Large-Scale Geographical Limitations: the use of national-level data may overlook specific local or regional differences, particularly in large countries, reducing the model’s precision. ⎫ Data heterogeneity and gaps: a lack of homogeneous and fine-scale spatial and social data limits the model’s ability to capture local AMR dynamics accurately.	⎫ Extend the model to sub-national levels using more granular indicators and homogeneous datasets to better capture local conditions. ⎫ Strengthen data collection through targeted campaigns and ensure predictions, especially those related to socio-environmental factors, are validated by domain experts to avoid misinterpretation and policy errors.
Neculai-Valeanu et al. [45]	Global	Review	To assess the growing issue of AMR on dairy farms and explore potential solutions through digital health monitoring and precision livestock farming.	Digital health	Animal health	⎫ Ongoing improvements in sensors, data analytics, and connectivity systems are progressively reducing implementation challenges and opening avenues for innovation in smart health. ⎫ Opportunities exist to foster partnerships between academia, industry stakeholders, and policymakers to accelerate adoption and address AMR.	⎫ Economic Constraints: high initial costs and limited financial capacity hinder the widespread adoption of technologies. ⎫ Data and Regulatory Challenges: Concerns around data privacy, interoperability, standardisation, and inconsistent regulatory frameworks across countries limit large-scale implementation.	⎫ Stakeholder engagement and policy support: all stakeholders must address technological, economic, and legal issues collaboratively, with policymakers playing a critical role in ensuring fair investments and sustainable agricultural practices.
Olatunji et al. [46]	Global	Review	To review the various AI methods and approaches for identifying and annotating ARGs, and highlighting their potential and limitations.	DL, ML	Not specified	⎫ AI can serve as a powerful assistive tool for domain experts in identifying and annotating ARGs, enhancing the efficiency of antibiotic resistance research. ⎫ Supervised learning and DL models show potential for improved ARG classification when combined with high-quality data and domain knowledge.	⎫ Supervised learning is limited by labelled data, meaning AI cannot identify ARGs outside the scope of its training labels. ⎫ Lack of high-quality, curated biomedical datasets and the need for computational resources (e.g., for assembly before ARG prediction) hinder AI’s broader application.	⎫ AI applications should be guided by domain experts to ensure meaningful interpretation and relevance of results in antibiotic resistance studies. ⎫ Investment in high-quality, well-annotated datasets is critical to improve AI model performance and generalizability in biomedical contexts.
Pascucci et al. [47]	France	Original article	An offline AI-based smartphone app for antibiogram analysis to support disk diffusion AST in resource-limited settings where routine use or interpretation is limited.	ML	Not specified	⎫ An AI-powered mobile app enables accurate, on-device analysis of disk diffusion antibiograms, with performance comparable to manual readings. It supports standard and enriched media, offers manual adjustment features, and improves AST access in low-resource settings. ⎫ The app also holds potential for contributing data to global AMR surveillance systems like WHO GLASS and WHONET.	⎫ Limited training data for ML models: the resistance mechanism detection models risk overfitting due to small datasets, limiting their reliability. ⎫ Pending clinical validation: the App’s effectiveness in real-world healthcare settings remains to be validated through further clinical trials.	Further real-world testing is needed to assess patient benefits and validate the App’s practical impact.
Pennisi et al. [5]	Global	Review	To explore the multifaceted role of AI models in enhancing antimicrobial stewardship efforts across healthcare systems.	ML	Human health	⎫ Enhanced decision-making and personalisation: AI and ML models enable precise, data-driven decision-making through predictive analytics, real-time monitoring, and personalised antibiotic therapy recommendations. ⎫ Improved Surveillance and Policy Support: AI-driven surveillance systems integrate diverse data (e.g., EHRs, lab results, environmental data) to detect AMR trends and provide actionable insights for healthcare providers and policymakers.	⎫ Data quality and Integration Issues: The effectiveness of AI tools depends on the availability and consistency of high-quality, standardised data across various sources. ⎫ Lack of transparency and ethical concerns: many AI/ML models lack interpretability (the “black-box” issue), raising concerns about algorithm transparency, fairness, and ethical use in healthcare.	⎫ Future research should prioritise developing AI systems that are explainable to support trust and informed decision-making among healthcare professionals. ⎫ Encourage interdisciplinary collaboration: successful and equitable implementation of AI in antimicrobial stewardship requires collaboration across clinical, technological, and ethical disciplines.
Rabaan et al. [6]	Global	Review	To evaluate the application of AI in combating high AMR rates	ML, artificial neural network.	Not specified	⎫ Enhance antibiotic stewardship by improving diagnosis, treatment accuracy, and drug discovery while reducing costs and time delays. ⎫ Support clinical decision-making, especially during emergencies where waiting for lab results is not feasible.	Limited infrastructure or awareness in developing countries may hinder the adoption and integration of AI tools into routine healthcare systems.	Capacity building and implementing institutional antibiotic stewardship programs.
Rodríguez-González et al. [48]	Global	Review	To provide an overview of the current application of AI in the field of public health and epidemiology, with a special focus on AMR.	ML	Environmental and human health	⎫ Innovative problem-solving potential: AI and ML offer radical new solutions to complex and persistent public health challenges such as AMR and climate change. ⎫ Access to diverse and rich data: increasing availability of real-world data, including from non-traditional sources like social networks and media, can enrich epidemiological studies.	⎫ The complexity of the integration of heterogeneous data, and the lack of sound and unbiased validation procedures. ⎫ Data accessibility and method transparency: limited access to quality data and the “black-box” nature of many AI/ML models hinder interpretability and trust in results.	⎫ Encourage the development of interpretable AI models that offer rational explanations for their outputs to improve trust and usability in public health.
Rusic et al. [49]	Global	Review	To provide an overview of available information about the application of ML techniques in AMR.	ML	Not specified	⎫ Accelerated scientific discovery: ML can significantly improve the speed and precision of predicting AMR, discovering new treatments, and optimising evidence-based decision-making. ⎫ AI can revolutionise biomedical research and healthcare by enabling personalised programming, efficient data analysis, and improved surveillance through objective learning algorithms.	⎫ Limited implementation despite high potential: there is a clear gap between the rapid development of ML tools and their practical application in real-world scientific and healthcare settings. ⎫ The success of ML models depends on large volumes of high-quality data, but current restrictions on data access and variability in data quality hinder progress.	⎫ Ensuring unrestricted, standardised access to large, high-quality datasets is crucial for advancing ML applications in AMR research. ⎫ Training researchers in ML, similar to statistical training, will empower more scientists to effectively apply AI tools in biomedical research.
Sakagianni et al. [50]	Global	Review	To discuss the applications of ML methods in the field of AMR and their value as a complementary tool in the antibiotic stewardship practice, mainly from the clinician’s point of view.	ML	Not specified	⎫ ML tools show promising potential for accurately predicting AMR, enabling more personalised antibiotic prescriptions and informed clinical decision-making. ⎫ Accessible Decision Support Systems: ML-based tools can assist healthcare providers, especially in low-resource settings, by offering rapid, data-driven support when diagnostic tests are unavailable or limited.	⎫ Data Quality and Generalizability Issues: ML models often rely on low-quality or non-standardised electronic health record (EHR) data, and their generalizability is limited. ⎫ Lack of ML Expertise Among Clinicians: A shortage of ML knowledge in the healthcare field hinders the deployment and effective integration of ML models into clinical workflows.	⎫ Promote external validation and real-world testing: ML models should be externally validated using diverse datasets and evaluated in real-world settings or randomised controlled trials (RCTs) to ensure clinical relevance and reliability. ⎫ Invest in interdisciplinary collaboration and training.
Shafiq et al. [51]	Global	Review	To outline how AI can improve AMR surveillance, analyse resistance trends, and enable early outbreak identification.	ML	Not specified	⎫ Enhanced AMR surveillance and response: AI can significantly improve AMR surveillance, predictive modelling, and early outbreak detection using large-scale clinical, environmental, and genomic data. ⎫ Personalised treatment and rapid diagnostics: AI can support personalised antimicrobial therapy and enable rapid identification of resistant strains through AI-based antimicrobial susceptibility testing (AST).	⎫ Data privacy, informed consent, and compliance with regulations, present major hurdles in using sensitive genetic and clinical data. ⎫ The lack of explainability (black-box models), heterogeneity of data, and inherent biases limit the reliability and generalizability of current ML models.	⎫ Develop explainable AI (XAI) and scalable models: to enhance trust and applicability, focus on refining models for transparency, accuracy, and adaptability across different settings. ⎫ Encourage partnerships among bioinformaticians, microbiologists, clinicians, and regulatory bodies to create effective, ethical, and tailored AMR interventions.
Steinkey et al. [52]	Canada	Original article	To assess the application of AI to the in-silico assessment of AMR risk to human and animal health	ML	Human and animal health	⎫ AI has already improved infectious disease identification and characterisation, the benefits of which will affect public health and animal health laboratories around the world. ⎫ In addition to AMR and bacterial populations of greatest risk to human health, AI algorithms hold promise as tools to predict other clinically and epidemiologically important phenotypes of enteric pathogens.	⎫ The size of dataset requited to effectively train ML models. ⎫ Lack of trained experts ⎫ Data quality and model validation issue.	Focus on improving data governance, model transparency, and validation frameworks to support trustworthy and responsible use of AI in human and animal health.
Sun et al. [53]	Global	Review	To provide a comprehensive summary and integration of AI to control or even eliminate the transfer of antibiotic resistance genes	ML /DL model	Agriculture	⎫ It helps the farmers and researchers to understand the resistance trends and identify the sources of contamination by advanced prediction and analysis ⎫ It helps target antibiotic use and minimises the development of ARG by enhancing monitoring and farm management and early warning.	⎫ Data heterogeneity, ⎫ Model interpretability, ⎫ Scalability across farm environments, ⎫ Data privacy and ethical use.	⎫ Given the complexity and diversity of ARGs sources and transmission pathways, in-depth research on ARGs control methods and interdisciplinary cooperation and integration will be crucial.
Zavaleta-Monestel et al. [54]	Global	Review	To analyse the usefulness of AI in antibiotic development, highlighting its benefits in terms of time, cost, and efficiency in the fight against resistant bacteria, as well as the challenges associated with its implementation.	Not specified	Not specified	⎫ Enhanced Treatment Options: AI can support the discovery of new antimicrobials and optimise combination therapies, particularly for treating infections caused by multidrug-resistant (MDR) bacteria in critical or immunocompromised patients. ⎫ Transformative Drug Discovery: AI reducing time and cost in drug discovery and offering solutions to manage and prevent outbreaks.	⎫ Effective AI implementation is hindered by the lack of high-quality. ⎫ Validation Challenges: AI-generated findings require rigorous experimental trials for clinical validation, which can delay implementation and reduce trust in AI-driven approaches.	⎫ Establish standardised, comprehensive, and high-quality data sources to improve AI model training and performance. ⎫ Integrate robust experimental and clinical validation processes to confirm the safety and efficacy of AI-predicted antimicrobial agents.

The role and opportunities of AI in combating AMR

The findings of the included studies regarding the opportunities of AI in combating AMR are summarised into eight main themes: (1) identification of resistant pathogens or resistance markers, (2) AI-powered analysis of clinical data, (3) AI-powered surveillance, monitoring and early warning, (4) integration and interoperability of heterogeneous data, (5) antimicrobial discovery and drug development, (6) optimising antibiotic stewardship and clinical decision support, (7) resource allocation and targeted interventions, and (8) support for evidence-based policy and public health decision-making.

Identification of resistant pathogens or resistance markers

AI has revolutionised AMR diagnostics by enabling rapid and precise identification of resistant pathogens. Studies demonstrated that ML models can analyse large-scale genomic clinical and laboratory data, including imaging and mass spectrometry, to detect resistance patterns much faster than traditional methods [17, 40, 42]. Importantly, Pascucci et al. [47] reported that the development of AI-powered diagnostic tools, such as the offline smartphone app, expands access to reliable antimicrobial susceptibility testing (AST) in resource-limited settings. Research by Kolluru et al. [37] and Monaco et al. [44] highlights AI’s potential to detect antimicrobial-resistant pathogens through advanced analysis of electronic health records (EHRS), genomic data, and medical imaging. Additionally, the included studies highlighted that AI models such as support vector machines, random forests, and neural networks can pinpoint genetic mutations and resistance markers linked to AMR, and of predict bacterial resistance by analysing gene content and genomic features [34, 44].

AI-powered analysis of clinical data

As demonstrated by Kolluru et al. [37], leveraging ML algorithms to extract meaningful patterns from large-scale electronic health records and patient datasets enables more accurate identification of infection sources, resistance trends, and patient risk profiles. According to Massé et al. [41] and Lau et al. [34], DL models such as convolutional neural networks (CNNs) can analyse whole-genome sequences or metagenomic reads to classify bacterial strains as resistant or susceptible by detecting single-nucleotide polymorphisms (SNPs) and resistance-related genetic elements.

AI-powered surveillance, monitoring, and early warning

AI-driven biosurveillance systems integrate and analyse large-scale, cross-sectoral data (human, animal, and environmental) in real time, providing early warnings, outbreak prediction, and efficient monitoring of AMR trends to support timely public health interventions and the One Health approach [43]. Most of the included studies discussed that AI-driven biosurveillance systems could transform how AMR trends are tracked and predicted across human, animal, and environmental health sectors. Studies by Sun et al. [53] and Ayesiga et al. [21] highlight that AI-driven surveillance and monitoring can process vast and diverse datasets in real time, providing early warnings for emerging resistance and helping identify sources of contamination. Two studies also revealed that the ability of AI to support real-time surveillance and early detection is particularly valuable for timely public health interventions [35, 42]. ML and DL have been used to map resistance dissemination pathways, as highlighted by Hatim et al. [35] and Olatunji et al. [46], These approaches serve as early warning systems by linking wastewater AMR profiles to human and animal sources, thereby enabling targeted interventions in high-risk communities. Studies have also shown that predictive analytics driven by AI can use both historical and real-time data to forecast the emergence and spread of AMR before outbreaks escalate. For instance, time series models like ML classifiers identify high-risk pathogens and geographic hotspots by analysing resistance trends, antibiotic usage patterns, and environmental factors [54, 55]. AI-driven surveillance systems continuously monitor genomic data to track the emergence and spread of resistance genes, enabling public health authorities to implement targeted interventions and containment strategies [28, 49].

Integration and interoperability of heterogeneous data

Studies conducted by Pascucci et al. [47], Steinkey et al. [52], and Zavaleta-Monestel et al. [54] emphasised that AI can be used to integrate biological, clinical, and environmental data to tailor antibiotic regimens for animals and humans. AI facilitates the aggregation and harmonisation of diverse datasets, including clinical, environmental, and epidemiological datainto unified platforms, enhancing interoperability, data sharing, and comprehensive AMR surveillance across sectors [24, 56]. Another study highlighted that AI facilitates the integration of heterogeneous data, including genomic sequences, clinical records, environmental samples, antimicrobial usage, and epidemiological data from human health, veterinary, agricultural, and environmental sources, thereby enabling unified platforms for comprehensive AMR surveillance in alignment with the One Health approach [2, 23]. These platforms enable scalable data storage, processing, and real-time analytics, facilitating cross-sector One Health approaches.

Antimicrobial discovery and drug development

In this study, almost all included articles discussed the application of AI in antibiotic discovery and drug development, particularly in accelerating the identification of novel antimicrobial agents and the repurposing of existing drugs, optimising molecular design and dosing, reducing drug development time and costs, and supporting more robust and sustainable pipelines. Reviews by Liu et al. [39] and Elalouf et al. [29] illustrate that ML models can screen large chemical libraries, guide molecular design, and predict structure-activity relationships and toxicity, thereby reducing development time and costs. A study by Anahtar et al. [20] and Abavisani et al. [16] shows that AI facilitates the discovery and design of new antibacterial drugs through predictive modelling and computational simulation, and potentially reduces associated costs. Gonzales et al. [55] and Mohammed et al. [2] also highlighted that AI can improve efficiency, accuracy, and speed in drug discovery processes by leveraging ML and DL for data analysis.

Optimising antibiotic stewardship and clinical decision support

Another important point reported by included studies was that AI plays a pivotal role in optimising antibiotic stewardship by supporting clinical decision-making, predicting treatment effectiveness, and improving prescribing practices [23, 34, 37, 50]. Studies such as Rabaan et al. [6] and Elyan et al. [30] show that AI-driven clinical decision support systems can assist clinicians in selecting the most appropriate antibiotics, especially in emergencies where rapid decisions are needed. By analysing large amounts of patient data, including genetic information, AI can aid in predicting individual responses to different antibiotics and tailor treatment plans to individual patients, thereby enhancing the effectiveness of therapies and reducing side effects [37]. Furthermore, studies showed that by integrating patient data, biomarkers, and local resistance patterns, AI models can guide individualised therapies and reduce unnecessary antibiotic use, ultimately supporting more effective stewardship programs and improving patient outcomes [3, 9, 24]. AI-driven antimicrobial stewardship programs (ASPs) evaluate prescribing outcomes and adherence to guidelines, and personalised antibiograms generated via ML improve dosing accuracy and minimise collateral damage to commensal microbiota [46].

Resource allocation and targeted interventions

The included studies emphasised that AI enhances the allocation of healthcare and agricultural resources by identifying high-risk populations and environments, supporting targeted interventions, and continuously assessing the effectiveness of stewardship and infection control measures [2, 27]. AI-driven insights enable healthcare systems, environmental, and agricultural stakeholders to allocate resources more efficiently by pinpointing where and when interventions are most needed [17, 27]. For instance, AI models can identify patient populations or farms at greatest risk for resistant infections, guiding focused antimicrobial stewardship programs and infection control measures.

Support for evidence-based policy and public health decision-making

Two included studies identified that AI supports evidence-based policy and public health decision-making for AMR by enabling real-time analysis of large and diverse datasets, forecasting resistance trends, guiding targeted interventions, and optimising resource allocation, thereby enhancing the effectiveness and adaptability of public health responses across the One Health sectors [38, 41].

Challenges and limitations of AI adoption for AMR

While AI presents groundbreaking opportunities for combating AMR through the One Health framework, its integration poses significant challenges and limitations. In review study, we identified various challenges and limitations and categorized them into six main themes: (1) data challenges and standardization; (2) model performance and generalizability; (3) interpretability, transparency, and trust; (4) ethical, legal, and privacy concerns; (5) infrastructure, resource, and capacity gaps; and (6) validation and real-world implementation.

Data challenges and standardisation

Included studies consistently report that AI models for AMR face major barriers due to data heterogeneity, as well as imbalanced or incomplete data, which undermine the reliability and generalizability of predictive algorithms [21, 36, 37, 39]. Ayesiga et al. [21] reported that the absence of standardised and high-quality data is a major barrier to developing scalable and reliable AI systems for AMR monitoring. Additionally, two studies pointed out that data quality and integration issues, including inconsistent data formats and incomplete records, present significant challenges for building effective AMR prediction models using AI [17, 46]. The included studies also emphasised that the primary challenges in integrating data for AMR surveillance are the fragmentation of data systems across human health, animal health, and environmental sectors [45, 53]. Each sector typically operates under distinct governance structures, regulatory frameworks, and institutional mandates, leading to siloed data collection and management. This fragmentation results in duplicated efforts, inconsistent data standards, and poor coordination, which hinder comprehensive surveillance and timely response to AMR threats [45].

Model interpretability, transparency, and trust

Several included studies point out that the “black-box” nature of advanced AI models, especially DL, makes it difficult for clinicians and stakeholders to interpret predictions, undermining trust and impeding adoption in clinical practice [29, 30, 33]. Another study also reported that the lack of model transparency and explainability is a significant barrier to regulatory approval and real-world integration [20, 35]. The complexity of integrating diverse data sources across human, animal, and environmental health, coupled with AI models’ black box nature, complicates understanding and trust among stakeholders [26, 32, 49].

Infrastructure, resource, and capacity gaps

Studies emphasise that limited digital infrastructure and a shortage of trained personnel, particularly in low-resource settings, are major obstacles to the widespread adoption and sustainability of AI-based AMR solutions [21, 28]. Lack of interdisciplinary expertise and training limits effective development, validation, and deployment of AI solutions [39]. Masud et al. [42] described high implementation costs and resource limitations as obstacles to AI adoption. In many low- and middle-income countries (LMICs), costs associated with data storage, processing power, and skilled personnel are significant barriers to implementation and sustainability [34].

Ethical, legal, and privacy concerns

Included studies raise concerns about data privacy, ethical use, and informed consent, especially when handling sensitive patient and genomic data in AI-driven AMR research and surveillance [6, 29]. Furthermore, studies also mentioned that unclear regulatory frameworks and the risk of algorithmic bias further complicate the ethical deployment of AI technologies [16, 20, 21]. Ensuring the confidentiality and integrity of health data while leveraging it for AI applications requires robust data protection measures, which can be complex and costly to implement [23, 26]. The study also highlighted that the lack of clear guidelines for AI use in AMR surveillance regarding ethical issues creates uncertainty for developers and healthcare providers [23].

Regulatory and policy gaps

Despite growing recognition of AI’s potential to enhance AMR surveillance and management, many countries face significant policy gaps that hinder its systematic adoption. The studies identified that variation in data governance and regulatory standards across countries impedes the global scalability and deployment of AI-driven AMR solutions [23, 42]. Additionally, the absence of unified data and model validation standards hinders cross-border collaboration and model deployment [3]. Abavisani et al. [16] raised that regulatory and policy issues, including unclear data governance frameworks and restrictive data-sharing policies across sectors and jurisdictions, impede collaboration, innovation, and effective implementation in AI-driven AMR solutions.

Model performance and generalizability

Studies highlight that while AI models can achieve high accuracy in controlled or internal datasets, their predictive performance often drops when applied to external datasets due to overfitting, concept drift, and lack of diverse training data [19, 33, 40]. This limited generalizability restricts the clinical and epidemiological utility of AI-driven AMR tools across different populations and healthcare settings [2, 31, 43]. Furthermore, studies also demonstrate that the lack of large, geographically and demographically diverse datasets hampers the development of models that can generalise across settings [19, 31]. Mohseni and Ghorbani [43] reported that limited access to high-quality training data and predictive accuracy are barriers to broader AI use.

Validation and real-world implementation

Some of the included studies noted that most AI models for AMR lack robust clinical validation and real-world testing, with few models evaluated outside of retrospective or laboratory settings, which limits their impact on patient care and antimicrobial stewardship [19, 26, 33]. There is a need for pragmatic clinical trials and real-world evidence to demonstrate the effectiveness and safety of AI-driven AMR interventions in practice [2, 50].

Discussion

In this scoping review, we have identified 43 studies addressing the opportunities and challenges of AI in combating AMR across the One Health sector. No grey literature was identified for inclusion. The review identified eight key opportunities for AI in mitigating AMR, as well as seven cross-cutting challenges that should be addressed for effective AI integration. The findings of our review indicate that the majority of publications are concentrated on applications of AI within human health. In contrast, only a few studies have reported on the application of AI in the contexts of animal health and the environment.

The findings underscore that AI can significantly advance the rapid identification of resistant pathogens and resistance markers, enhance clinical and genomic data analysis, and enable real-time surveillance and early warning systems by integrating heterogeneous data from human, animal, and environmental sources [31, 46, 57]. Notably, the development of AI-powered diagnostic tools, such as smartphone-based applications, has expanded access to reliable antimicrobial susceptibility testing, particularly in resource-limited settings [47]. AI-powered analysis of clinical data enables the extraction of meaningful patterns from vast electronic health records and patient datasets, facilitating and supporting more precise antibiotic stewardship, facilitating novel antimicrobial discovery and drug development, and improving resource allocation and targeted interventions [44]. Furthermore, DL models, such as CNNs, have further enhanced the classification of bacterial strains and the detection of resistance-related genetic elements [41]. For instance, ML models including XGBoost and convolutional neural networks, have been used to predict AMR in Klebsiella pneumoniae, accurately identifying strains with high drug resistance or virulence from genomic and clinical data much faster than traditional laboratory techniques [58].

Another finding of our results is that most of the current data analysis is limited mainly to individual sectors. However, AMR is a critical global health threat that requires scalable solutions integrating AI within a One Health framework, which recognises the interconnectedness of human, animal, and environmental health [9, 10, 59]. Our finding indicates that AI-driven biosurveillance systems integrate and analyse cross-sectoral data from human, animal, and environmental sources in real time, providing early warnings, outbreak predictions, and efficient monitoring of AMR trends [22, 23]. This can support public health decision-making and timely interventions by providing actionable insights from large and complex datasets. By enabling comprehensive surveillance across the One Health spectrum, these AI-powered systems facilitate more coordinated and proactive responses to emerging resistance patterns, ultimately reducing the risk of widespread AMR outbreaks and improving population health outcomes. We also found that AI can facilitate the integration and interoperability of heterogeneous data, aggregating and harmonising clinical, environmental, and epidemiological datasets into unified platforms for comprehensive AMR surveillance [23, 24, 30, 47]. This supports scalable data storage, real-time analytics, and cross-sector collaboration, essential for effective One Health strategies, and enables real-time monitoring and informed decision-making at both local and global levels [5].

Another important opportunity of AI lies in antimicrobial discovery and drug development, where it facilitates the identification of novel agents, optimises molecular design, and reduces development time and costs [2, 16]. For instance, ML and DL models can efficiently screen chemical libraries, predict structure-activity relationships, and guide the repurposing of existing drugs, thereby supporting robust and sustainable drug pipelines. An example supporting our discussion is the landmark study by MIT researcher who used a ML algorithm to identify a novel antibiotic called halicin. This AI-driven approach screened over 100 million chemical compounds in just days, discovering halicin’s unique ability to kill a broad range of problematic bacteria, including those resistant to all known antibiotics [60]. This demonstrates that AI is a powerful tool for strengthening and sustaining the antimicrobial drug pipeline, which is crucial for addressing the growing challenge of AMR. Optimising antibiotic stewardship and clinical decision support is another critical area where AI excels. AI-driven systems assist clinicians in selecting appropriate antibiotics, predicting treatment effectiveness, and tailoring therapies to individual patients [5, 27, 61]. This would support antimicrobial stewardship programs by evaluating prescribing practices and generating personalised antibiograms.

AI also enhances resource allocation and targeted interventions by identifying high-risk populations and settings, supporting focused stewardship programs, and dynamically evaluating the effectiveness of infection control measures [27, 62]. A study conducted by Botha et al. [63] highlighted that the positive impact of AI tools on patient care and confidentiality, documenting improved healthcare delivery and data protection within clinical settings. Studies demonstrate that AI algorithms, including decision trees, random forests, and neural networks, have been successfully applied to analyse diverse healthcare and environmental datasets to identify high-risk patient populations and settings prone to AMR [13, 21]. For instance, AI-driven models have been used to optimise antimicrobial stewardship programs by predicting where resistant infections are most likely to occur, enabling tailored interventions that improve patient outcomes and resource utilisation [13]. This would allow for the rapid analysis of diverse healthcare and environmental data to pinpoint where interventions are most needed, optimise the use of limited resources, and adapt strategies dynamically as new information emerges, ultimately enhancing patient outcomes and strengthening public health responses to AMR threats [64]. Furthermore, AI supports evidence-based policy and public health decision-making by enabling real-time analysis of large and diverse datasets, forecasting resistance trends, and guiding targeted interventions across the One Health sectors. To facilitate effective surveillance, One Health governance systems must include the political sphere to reconcile competing perspectives and interests by overseeing the coordination of legislation, policies and programs, knowledge, and resources across the human, animal, and environmental health sectors [65]. Such frameworks facilitate harmonised policies, reduce duplication, and ensure sector accountability.

Despite these opportunities, AI integration in AMR monitoring and management faces substantial challenges that should be addressed to realise its full potential. Our review indicates that data heterogeneity, lack of standardisation, and fragmented data systems across human, animal, and environmental health sectors undermine the reliability and scalability of AI models for AMR surveillance and prediction. This barrier makes it difficult for AI algorithms to integrate and analyse information across sectors, reducing reliability, limiting scalability, and posing challenges in generalising AI models to new settings or populations [66]. As a result, these issues pose significant obstacles to developing robust, accurate, and widely applicable AI-driven solutions for AMR monitoring and intervention. For example, Arnold et al. [3] describe that fragmented data systems across human, animal, and environmental health sectors severely undermine the reliability and scalability of AI models. Additionally, Li et al. [67] emphasise that the lack of comprehensive, harmonised datasets hampers the development of widely applicable AI-driven AMR surveillance systems, posing significant barriers to building accurate and scalable prediction tools. Furthermore, model interpretability, transparency, and trust remain significant barriers [36, 46]. The black-box nature of advanced AI models complicates the interpretation of predictions, undermining clinician and stakeholder confidence and impeding regulatory approval and real-world integration. As a result, this undermines trust in AI-driven decisions, creates hesitation in clinical adoption, and poses obstacles for regulatory approval [67]. Additionally, the challenge of integrating diverse data sources from these sectors adds further complexity, making it even harder to build transparent and trustworthy AI systems for AMR management.

Another important point we found in this review is that infrastructure, resource, and capacity gaps, particularly in LMIC, limit the widespread adoption and sustainability of AI-based AMR solutions [28]. This is because AI requires high implementation costs, advanced digital infrastructure, and well-trained personnel, which are often lacking in developing countries. f there are no intentional investments, low- and middle-income countries (LMICs) may not be able to take full advantage of artificial intelligence (AI) for antimicrobial resistance (AMR) surveillance. As a result, they face the risk of falling behind, which could worsen existing inequalities in how the world responds to global health challenges. Furthermore, ethical, legal, and privacy concerns, such as data protection, informed consent, and algorithmic bias, remain major obstacles, compounded by unclear regulatory frameworks and the lack of clear guidelines for responsible AI use [23, 29]. The use of AI in healthcare poses risk of unpredictable errors, regulatory gaps, increased costs, data privacy issues and potential bias, raising concerns about medical paternalism and insurance disparities [68]. The absence of clear regulatory frameworks and guidelines for AI in AMR monitoring raises risks of algorithmic bias (e.g., models trained on non-representative data failing in LMIC contexts) and data exploitation (e.g., misuse of sensitive health data from vulnerable populations) [69]. For example, reviews note that the lack of clear regulatory frameworks and guidelines for AI applications in AMR monitoring increases the risk that algorithms trained primarily on high-income country data may not perform reliably when applied in LMICs, thus perpetuating bias and reducing equity in health outcomes [70]. This is further complicated by fragmented and heterogeneous data systems across different sectors, which make it challenging for AI models to integrate, generalise and scale effectively for global AMR surveillance. Moreover, while AI models may perform well on internal datasets, their generalizability is often limited due to overfitting, lack of diverse training data, and concept drift, restricting their broader clinical and epidemiological [32, 33]. As a result, these limitations restrict the broader clinical and epidemiological applicability of AI models, meaning they may not be reliable or effective when used across different healthcare settings or populations. This underscores the need for diverse, representative datasets and ongoing model validation to ensure AI tools are robust and generalizable. Addressing these challenges requires a concerted effort from researchers, healthcare professionals, policymakers, and technology developers. By developing clear ethical guidelines, robust data governance frameworks, and interoperable healthcare technologies, we can navigate these obstacles.

Future directions, research needs and recommendations

Despite the considerable promise AI in advancing AMR surveillance and management within the One Health framework, this review highlights several critical areas for future research and policy action.

Strengthen One Health integration

Current AI applications mostly focus on human health, with limited attention to animal and environmental sectors. Future research should prioritise cross-sectoral AI models that integrate human, animal, and environmental data to align with the One Health framework.

Improve data quality, standardisation, and interoperability

Heterogeneous, fragmented, and incomplete datasets remain a critical barrier. To address these issues, future efforts should focus on creating standardised data formats, harmonised protocols, and interoperable platforms that facilitate seamless data integration across sectors and regions [24, 30]. Collaborative initiatives among stakeholders-including governments, healthcare providers, researchers, and industry partners-are essential to develop and implement these standards.

Address ethical, legal, and policy gaps

Current challenges include data privacy, informed consent, algorithmic transparency, and liability, all of which require urgent attention as AI adoption accelerates [29]. Research needs include establishing clear, context-appropriate guidelines for data governance, privacy protection, and equitable access, especially considering the underrepresentation of certain populations and regions in training datasets.

Development of explainable AI models

Explainable AI (XAI) models are crucial for AMR applications to ensure transparency, interpretability, and trust among different sectors. Unlike “black-box” models, XAI provides clear insights into how predictions or recommendations are made, enabling users to understand the rationale behind AI-driven decisions such as antibiotic prescribing or resistance risk assessment. This interpretability is essential for clinical adoption, regulatory approval, and ethical accountability.

Capacity building

To unlock the full potential of AI in AMR surveillance and management, it is essential to develop interdisciplinary expertise that encompasses environmental health, veterinary medicine, and public health. Strengthening capacity in these areas is especially important for LMICs, where limitations in skilled personnel currently hinder the effective implementation of AI technologies.

Limitation

Although our scoping review has strengths, it also has some limitations. The majority of the included studies focused primarily on human health, while significantly underrepresenting the application of AI in animal health and environmental contexts. As a result, the review may not fully capture the potential or specific requirements for AI integration in these sectors, which are critical for comprehensive AMR surveillance and management. Moreover, restricting inclusion to English-language publications may have introduced language bias and excluded relevant studies in other languages. Furthermore, by including studies addressing One Health perspectives, even if not as a primary focus, the review may encompass research with varying degrees of relevance and integration.

Conclusion

AI presents groundbreaking opportunities to transform AMR surveillance, diagnosis, stewardship, and policymaking under the One Health framework. AI-driven biosurveillance platforms aggregate and analyse data from human, animal, and environmental sources, delivering real-time alerts and outbreak predictions. However, addressing the technical, ethical, and systemic challenges identified in this review is crucial to ensuring the effective, equitable, and sustainable integration of AI into global AMR strategies. Continued interdisciplinary collaboration and policy innovation will be essential to harnessing AI’s full potential in the fight against AMR. This requires investment in explainable AI, better data infrastructure, stronger cross-sector collaboration, and clear regulatory frameworks to ensure ethical and effective use within the One Health approach.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

AI

Artificial intelligence

AMR

Antimicrobial resistance

CNN

Convolutional neural networks

DL

Deep learning

FAO

Food & agriculture organisation

LMIC

Low- and middle-income countries

ML

Machine learning

SNP

Single-nucleotide polymorphisms

UNEP

United Nations environment program

WHO

World health organisation

WOAH

World organisation for animal health

Author contributions

Gashaw Enbiyale Kasse: Conceptualisation, Data curation, Formal analysis, Methodology, Software, Writing-original draft, Writing-review and editing. Suzanne M. Cosh: Conceptualisation, Investigation, methodology, Supervision, Visualisation, Writing-review and editing. Judy Humphries: Conceptualisation, Investigation, methodology, Supervision, Visualisation, Writing-review and editing. Md Shahidul Islam: Conceptualisation, Investigation, methodology, Supervision, Visualisation, Writing-review and editing.

Funding

This review did not receive any financial support.

Data availability

All data generated or analysed during this review are included in this manuscript.

Declarations

Ethical approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.