Exploring perceptions of data risks in AI-enabled nursing research: A qualitative study

Background

Artificial Intelligence (AI) is an algorithmic system that learns from large-scale data and performs inference, prediction, and decision support and it is part of the broad reality of Digital Health and health systems development^1–5 With the data-driven transformation of the healthcare field and the continuous improvement of nursing information systems, AI enhances nursing efficiency, predicts nursing outcomes, and supports decision-making by analyzing large amounts of nursing data.^6–8 Machine learning-based nursing sensitive outcome prediction models have demonstrated powerful predictive capabilities and dynamic monitoring advantages.^9,10 These applications have not only improved the efficiency and accuracy of nursing work but also had a profound impact on nursing research.

However, some studies have pointed out the shortcomings of artificial Intelligence in nursing and medical management applications. Research by Chen. ¹¹ shows that AI does not adequately consider the needs of specific nursing scenarios, and its discussion of data privacy, security, acceptability, and ethical, legal, and social issues is limited. Katherine ¹² argues that a digital divide exists between the accessibility and application of AI tools and that some AI tools are not fully adapted to actual nursing workflows. Currently, insufficient technology adaptability and a lack of risk control remain pressing problems that need to be addressed.^7,13–15

Research by Sobhia et al. ¹⁶ highlights that the primary data risks associated with AI applications in the healthcare field include privacy breaches, insufficient data authenticity, algorithmic bias and lack of fairness, inadequate model interpretability, and unclear attribution of responsibility. In her review of AI ethics research, Adrianna ¹⁷ proposed that the ethical risks of AI in the healthcare field can be correlated with the core principles of “autonomy, authenticity, non-harm, and impartiality.” She emphasized that these risks stem not only from the technology itself but also from data governance systems, the definition of institutional responsibility, and user behavior norms. Özbay et al. ¹⁸ also believed that ethical concerns, infrastructure limitations and insufficient training of artificial intelligence restrict its effective integration and application. The application of AI is not merely a technological issue but also relates to the core human-centered philosophy of nursing. ¹⁷ Unlike highly structured data such as medical images and laboratory test results, nursing data exhibits a distinct characteristic—nursing assessments often involve substantial subjective judgment and unstructured records. It is precisely this unique attribute of nursing that makes the risks associated with AI in nursing research even more complex.

In the context of nursing, the risks associated with AI-related data exhibit strong context-dependent and discipline-specific characteristics. Nursing research data often originates from real clinical care scenarios and frequently contains sensitive information, such as patient identification and nursing assessment records. If researchers lack awareness of data anonymization, access control, and compliance when using AI tools for text processing, data analysis, or paper writing assistance, the risk of privacy breaches and unauthorized sharing may increase. ¹⁹ Nursing data contains a high proportion of unstructured text, subjective judgment indicators, and dynamic assessment content, and discrepancies in data collection and recording standards further hinder data processing. The study by Hongshin Ju et al. ²⁰ found that AI-generated nursing records suffer from a lack of specific details, inappropriate use of medical terminology, and an overall textbook-like, rigid style. Nursing research emphasizes the interpretability and clinical usability of evidence. If AI models rely excessively on black-box algorithms and lack traceable explanation mechanisms, it will not only weaken clinical nurses’ trust in research evidence but also undermine their confidence in AI models. However, it may also reduce the feasibility of translating research findings into clinical practice. ²¹ Nursing research often involves special populations such as the elderly, children, and patients with chronic diseases. If models are biased or yield misleading conclusions, this can lead to unfair interventions or harm to specific populations, thereby fostering more complex ethical and safety consequences in nursing research. ²²

It is noteworthy that recent research in nursing education has increasingly focused on the attitudes, literacy levels, and ethical understanding of nursing teachers and students towards AI technology.^23–26 Sarıkahya et al. ²⁷ showed that nursing teachers believe AI tools have the potential to enhance nursing education by supporting theoretical learning, improving efficiency, and promoting personalized learning experiences. At the same time, Research by Heather et al. ²⁴ suggests that undergraduate nursing students recognize the potential of AI to enhance work efficiency; however, they also worry about the risks of over-reliance on artificial intelligence and the dangers of a lack of regulation. Joyce et al. ²⁸ suggest that AI writing tools can help graduate nursing students enhance their writing skills and achieve academic success; however, a balance must be struck between AI use and the cultivation of independent critical thinking, while maintaining academic integrity. Existing research suggests that generative AI can enhance learning efficiency and care plan generation; however, it also raises concerns, including academic integrity issues, weakened critical thinking, and data privacy risks. These problems not only affect the quality of education but may also extend to the research training stage, potentially impacting the research literacy of nursing graduate students.^29,30

Overall, existing research still has three shortcomings: First, nursing AI research primarily focuses on model predictive efficacy and clinical application prospects, with less systematic discussion from the perspective of nursing research practitioners regarding the data risk perception structure and behavioral mechanisms in AI use ³¹ ); second, research on nursing graduate students often remains at the level of attitudes, acceptance, or user experience, lacking empirical evidence on risk identification capabilities, understanding of usage boundaries, and implementation of prevention and control measures ³² ; third, existing training and guidance often emphasize technical operations or principle explanations, lacking a risk identification and response support framework covering the entire nursing research process. ³³

Nursing graduate students play a crucial role in the standardization and advancement of high-quality nursing research. Their risk perception and response capabilities are not only related to research compliance and academic integrity, but also affect the reliability of research results and their clinical translational value. Therefore, it is necessary to analyze the characteristics, formation mechanisms, and support needs of data risk perception during AI use from the perspective of nursing graduate students, providing a basis for constructing AI use norms and educational intervention strategies that conform to the characteristics of the nursing discipline.

Given the above shortcomings, this study employs a qualitative research method to provide an in-depth description of the subjective experience of “risk perception.” Compared with quantitative surveys, interviews can more fully reveal how nursing graduate students understand data risks in the real context of AI-assisted research, how they weigh convenience and compliance, and their concerns and confusions in the “use-generation-sharing-responsibility” chain, providing more context-explanatory evidence for the subsequent development of targeted training strategies and governance frameworks.

Research objectives and questions

This study aims to explore the perceptual structure and connotation of data-related risks among nursing master’s students during the process of using artificial Intelligence to empower nursing research, and to propose suggestions for research management and postgraduate training based on this. The research questions are: ① What data risks do nursing postgraduate students mainly focus on throughout the entire process of AI-assisted research? ② How are these risks understood and expressed?

Research design and theoretical framework

This study employs a qualitative theme analysis approach, using original interview data as the basis for analysis. Through a systematic coding and extraction process, it identifies, analyzes, and interprets recurring core themes and patterns in the data, ultimately constructing an analytical framework that provides a deep explanation of the research phenomenon. The research team, research environment, data collection, and analysis processes will be presented clearly and transparently to enhance research transparency and traceability.

Research team and reflective approach

Research subjects and recruitment

Inclusion and exclusion criteria

Sample size determination

This study employed a qualitative topic analysis approach, with data saturation as the primary criterion for determining sample size. A small purposive maximum difference sample was selected to conduct exploratory data collection and initial coding. As the study progressed, the sample size was dynamically adjusted, and new sample data were repeatedly compared with the extracted themes and coding results using a continuous comparison method. Data saturation was achieved when no new initial codes or secondary themes emerged from 3–4 consecutive participants, and the existing data sufficiently supported the extraction of the core themes, at which point sample collection ceased. The final sample size was set at 20 participants. During recruitment, 22 potential respondents were contacted, but two were unable to participate in the formal interviews due to scheduling conflicts. No participants withdrew during the formal interview phase. In the event of withdrawal, the research team will maintain the same confidentiality regarding the information provided, in accordance with the informed consent agreement, and will not pursue further follow-up.

General information of respondents

Of the 20 respondents, 16 were female and four were male; aged 22-26; 5 were first-year students, 11 were second-year students, and 4 were third-year students; 17 were professional Master’s students and 3 were academic Master’s students; their research interests covered 12 fields including geriatric nursing, obstetric nursing, and chronic disease nursing; and they had been using AI for 2-4 years. See Table 1 for details.

OPEN IN VIEWER

Table 1.

General information of interviewees (n=20).

ID number	Gender	Age (years)	Grade	Degree category	Research direction	Duration of artificial intelligence use (years)	Institution
A1	female	25	2	Professional	Geriatric Nursing	3	C
A2	female	22	1	Professional	Obstetric Nursing	2	C
A3	female	23	2	Academic	Chronic Disease Nursing	3	C
A4	Male	23	3	Professional	Weight Management	3	A
A5	female	24	2	Professional	Pediatric Nursing	3	A
A6	female	23	1	Academic	Oncology Nursing	2	A
A7	Male	24	2	Professional	Bone and Joint Nursing	3	A
A8	female	23	1	Professional	Neonatal Nursing	2	C
A9	female	26	2	Professional	Health Care Nursing	3	C
A10	female	25	3	Academic	Neurological Nursing	2	A
A11	Male	25	2	Professional	Chronic Disease Nursing	2	B
A12	female	23	2	Professional	Community Nursing	3	C
A13	female	26	2	Professional	Psychiatric Nursing	3	C
A14	Male	22	1	Professional	Adult Nursing	2	C
A15	female	23	2	Academic	Geriatric Nursing	2	B
A16	female	26	3	Professional	Obstetric Nursing	4	B
A17	Male	24	2	Professional	Health Care Nursing	3	C
A18	Male	23	1	Professional	Rehabilitation Nursing	2	B
A19	female	25	3	Professional	Chronic Disease Nursing	3	C
A20	female	24	2	Professional	Chronic Disease Nursing	2	C

Data collection

Based on the research objectives and literature review, a preliminary analytical framework was constructed, and four nursing master’s students were selected for pre-interviews. One-on-one online interviews were conducted, each lasting 20-30 minutes, with the entire interview recorded and transcribed within 24 hours. After refining the framework based on feedback from the pre-interviews, two nursing research experts and one qualitative methodology expert were invited to review its effectiveness, ultimately establishing the formal interview framework.

Data analysis

Coding process

This study strictly followed the three-level coding process of classical grounded theory for data analysis. Researchers meticulously read the transcribed text line by line, extracting initial concepts related to data risk perception, and categorizing and integrating semantically similar concepts to form preliminary categories. The study employed a continuous comparative approach, repeatedly comparing relationships across categories to clarify their inherent logical connections and gradually establish a relatively systematic system of main and subcategories. Core categories were extracted from the main categories, and a logically clear “storyline” was constructed to connect the categories organically, ultimately forming a holistic theoretical framework. Two researchers independently coded and then compared and discussed their results. If coding disagreements arose, a consensus was reached through team consultation, and qualitative research experts were invited to participate in the review. 15%–20% of the transcribed text was randomly selected and compared with the original recordings to ensure the accuracy of the data source and the reliability of the analysis. To improve analytical transparency, a three-level structure of “initial concept—aggregated concept—category” was established during the open coding stage, and a condition-action-outcome relationship chain was established during the central axis coding stage around “AI-enabled research process (use—generation—sharing—responsibility)”. The final coding tree comprises five main themes and 20 sub-themes (see Table 2), retaining the original sentence index of key concepts to form a traceable “concept-evidence” correspondence table.

OPEN IN VIEWER

Table 2.

Themes and subthemes.

Themes and Subthemes	​
Theme 1: Data Adaptation Risks	1.1 Insufficient adaptability to professional scenarios
	1.2 Difficulty in processing unstructured data
	1.3 Fragmented AI model system
Theme 2: Data Security Risks	2.1 Leakage of research information
	2.2 Leakage of sensitive information
	2.3 Blurred boundaries of use
	2.4 Misuse of research data
	2.5 Intellectual property infringement
theme 3: Data Quality Risks	3.1 Fabrication of Professional Information
	3.2 Data Output Bias
	3.3 Bias Towards Conformity to Results
	3.4 Amplification of Data Bias
theme 4: Data Ethics Risks	4.1 Bias in the perception of responsibility
	4.2 Unclear ethical boundaries
Theme 5: Related Response Risks	5.1 Lack of risk awareness
	5.2 Widespread lack of protective measures
	5.3 Lack of training on usage guidelines
	5.4 Inadequate legal constraints
	5.5 Insufficient platform security protection
	5.6 Lack of research guidelines

Research rigor, credibility, and ethical considerations

Core risk themes and sub-themes

The structural contradiction between the professionalism of nursing research and the generality of AI technology

This study shows that the usability of current AI tools in the nursing professional context is limited. Artificial intelligence has shown great potential in the medical field due to its advantages in efficient data processing, pattern recognition and automation. However, the essential attributes of nursing are misaligned with the logic of existing AI technology, which limits the deep application of AI in nursing practice. When faced with a large amount of unstructured nursing research data, general models exhibit comprehension bias and unstable information extraction. The root of this bias lies in the fact that the worldview of general AI is built on massive amounts of general-language data, lacking an internalized understanding of nursing’s unique knowledge system. Its algorithmic logic focuses on statistical correlation patterns, while nursing professional thinking integrates practical wisdom that combines scientific evidence, clinical experience, ethical judgment, and humanistic care. This contradiction makes general models prone to semantic loss and context stripping when dealing with nursing professional issues, resulting in outputs that lack professional depth and accuracy. Kwan et al. ³⁵ also noted that the nursing field contains a large amount of unstructured text data, which often needs to be placed in a specific clinical context to be accurately interpreted. Especially in high-risk scenarios such as emergency and psychiatric care, AI output needs to provide clear evidence to establish clinical trust.^36,37 However, some studies hold the opposite view, arguing that, through natural language processing, AI systems can effectively analyze unstructured text in nursing records and identify subtle changes in patients’ emotions or early signals of potential complications.^38–40 Overall, the training datasets of current mainstream artificial intelligence models have not fully captured real-world nursing scenarios, making it difficult for these models to effectively identify and extract core professional information in nursing practice. This has led nursing researchers to repeatedly try multiple artificial intelligence tools when dealing with data closely related to clinical decision-making. Morey et al. ⁴¹ further emphasized that in clinical decision support, when AI recommendations deviate from the actual situation, they may interfere with nurses’ judgment. These phenomena reflect the nursing field’s expectations for artificial intelligence tools: they not only need to be efficient and fast, but also need to truly understand the inherent clinical logic and humanistic care of nursing work. ⁴²

Existing research has proposed relatively clear directions for improvement. Ruan et al. ⁴³ believe that the nursing professional knowledge system should be deeply embedded into model training, rather than relying solely on general corpora. Martin et al. ⁴⁴ explored the potential of introducing emerging technologies, such as multimodal large language models, into nursing practice in a review, emphasizing their potential to reduce paperwork burden and improve clinical efficiency. Multiple studies^13,45,46 have consistently shown that to effectively promote the integration of AI into nursing practice, nurses must be involved from the early stages of development, incorporating nursing professional judgment into the selection of training data, the verification of clinical scenarios, and the formulation of assessment standards. Only by integrating professional judgment into the underlying technology can the “nursing applicability” of the model be fundamentally improved. ⁴⁷ Therefore, nursing education should proactively promote interdisciplinary integration, foster deep collaboration among nursing, computer science, linguistics, and other fields, and cultivate talents with both nursing professional and technical application skills. Courses such as nursing informatics and health data science can be added to cultivate nursing students’ data literacy and critical technology thinking. By leveraging real-world nursing data to train in AI-assisted decision-making, students can improve their ability to identify and evaluate AI suggestions, while emphasizing that AI should serve only as a decision-making aid for nurses, not a replacement. Nursing experts will lead the optimization of workflows, seamlessly integrating AI tools into daily nursing work and clearly defining responsibilities. Simultaneously, they can lead the development of AI models specifically designed for nursing, using standardized nursing data for professional training, allowing the technology to better adapt to actual clinical nursing needs.

The dual challenges of sensitive data protection and AI technology characteristics

The risks of research information leakage, sensitive data breaches, and data misuse identified by respondents in this study are highly consistent with international discussions on the risks of AI applications in healthcare. Artificial intelligence essentially relies on large amounts of data and pattern recognition to operate. In the process of continuously optimizing performance, it is very likely to cross the bottom line of privacy protection inadvertently. Even anonymized data may still be re-identified through model memory and reverse reasoning. ⁴⁸ Although multi-center data aggregation can expand the sample size, it also creates a more concentrated pool of sensitive data, which is more harmful if leaked. ⁴⁹ If the training data itself has historical biases, AI may also reinforce and amplify them, causing group injustice in the allocation of nursing resources and elevating the problem from a privacy issue to a social fairness issue. ⁵⁰ Nursing plans generated by large language models have been proven to have “subtle but reproducible biases”, which may exacerbate inequality in the allocation of medical resources. ⁵¹ In studies of diseases such as cardiovascular disease and cancer, if group biases related to demographic variables are not corrected, they will directly affect the fairness of clinical decision-making. ⁵² Faced with the dual challenges of data security and the application of technology, the nursing field needs to build an active protection and adaptation system at the educational and clinical levels. In nursing education, health data ethics courses should be integrated, and case-based teaching should be conducted through real-world AI scenarios to cultivate students’ data ethics judgment. ⁵³ In clinical practice, it is necessary to establish a privacy-first intelligent application environment, conduct ethical and safety assessments before introducing AI tools, actively understand and pilot cutting-edge privacy protection technologies such as federated learning and synthetic data generation, strengthen patients’ right to know and participation, and have nurses do a good job in communication and rights protection, and use intelligent technologies safely under the premise of transparency and compliance.^54,55

The risks identified in this study are dual and overlapping: they involve both the leakage of sensitive patient information and the protection of researchers’ academic intellectual property. A systematic review notes that privacy leakage is the most prominent risk of medical artificial intelligence, primarily manifested in the disclosure of sensitive patient information and the theft of researchers’ intellectual property. ⁵⁶ This is closely related to the lack of transparency in the data storage and reuse process of artificial intelligence. ⁵⁷ Nursing research includes both patient data and unpublished research ideas, pathways and interim results, thus facing a dual risk of leakage. However, the risks caused by the blurred boundaries of use have not received sufficient attention. Students are generally concerned that artificial intelligence may undermine academic integrity, so it is essential to clarify its boundaries of use. ²⁸ The reuse of internal data by artificial intelligence systems and the dissemination of information across platforms are directly related to their data storage mechanisms. Researchers’ insufficient understanding of the mechanisms underlying artificial intelligence data processing will significantly exacerbate security risks. ⁵⁸ In this study, nursing graduate students have limited knowledge of relevant rules, which amplifies the potential risks of infringement and leakage. Content generated by artificial intelligence is very likely to cause serious copyright ownership disputes and may escalate into academic misconduct disputes. ⁵⁹ Therefore, it is necessary to clarify the auxiliary role of artificial intelligence in nursing research, strictly define its application scope, and avoid the decline in research rigor due to over-reliance on technology. ⁶⁰ In the face of the above risks, it is imperative to build a systematic governance framework. The comprehensive governance model, based on risk classification, established by the EU Artificial Intelligence Act, provides an important reference for policy design in nursing. ⁶¹ The Act emphasizes technological transparency and accountability mechanisms and, through regulation, promotes the efficient use of artificial intelligence while effectively controlling the risk of information leakage, providing a valuable framework for the standardized application of artificial intelligence in nursing research.

The core tension between AI output objectivity and research rigor

The risks of research information leakage, sensitive data leaks, and data misuse identified by respondents in this study are consistent with those associated with AI applications in international healthcare. The respondents’ concerns that AI could amplify inherent social biases in data are consistent with the findings of Andrew et al. ⁶² The core contradiction in the application of artificial intelligence in nursing lies in the inherent conflict between its technical objectivity and the requirements of research rigor. The uninterpretable nature of the decision-making process of complex models contradicts the requirements of nursing research for logical clarity and rigorous causality. ⁶³ At the same time, fixed models trained on historical data are difficult to adapt to the dynamic, highly individualized and complex situations in clinical practice. ⁶⁴ The research by Rengers et al. ⁶⁵ further confirms that large language models may exhibit a bias toward specific results when generating content. The conclusions produced by AI systems appear objective and fair, but in reality, they are deeply dependent on data quality, labeling subjectivity, and algorithm design. Their black-box nature also fundamentally contradicts the principles of traceability and verifiability required by scientific research. ⁶⁶ This contradiction is particularly prominent in nursing research evaluating the effects of clinical interventions, because artificial intelligence tends to generate answers that conform to the user’s pre-set ideas. Such bias may potentially affect the safety of nursing practice. ⁶⁷ In addition, the respondents’ concerns about artificial intelligence providing false data and literature echoed the findings of Heather et al., ²³ which found that artificial intelligence does indeed create illusions in academic writing, including falsifying references and tampering with data.

Faced with the above challenges, the nursing field needs to build a critical integration of artificial intelligence capabilities in education and practice. ⁶⁸ In education, research methods and AI literacy should be integrated into teaching, and content related to health AI assessment should be offered to cultivate students’ critical thinking ability. Through interdisciplinary collaborative learning, AI models should be made more interpretable and verifiable. ⁶⁹ In clinical practice, the auxiliary research role of AI should be clarified, and its results should be verified through scientific research and clinical judgment before they are used. A localized verification and routine monitoring mechanism for AI models should be established, and transparent reporting standards should be followed in research publications to form a rigorous human-machine collaborative closed loop and ensure the scientificity and reliability of artificial intelligence applications.^70,71 In terms of academic integrity, it is particularly important to establish an effective mechanism for identifying and verifying AI-generated content. The study by Miriam et al. ⁷² pointed out that the frequency of use of specific conjunctions and high-frequency words can serve as clues to distinguish between original writing by nursing students and AI-generated text. This suggests that the nursing research field needs to establish a more targeted AI output verification mechanism that focuses on the accuracy of terminology, the verifiability of cited literature, the reliability of data sources, the representativeness of samples, and the absence of jumps in the reasoning chain. Only by constructing a multi-dimensional and structured verification framework can we leverage artificial intelligence to improve research efficiency while upholding the bottom line of rigor and authenticity in nursing academic research.

The dual deficiencies in responsibility division and ethical norms

In the context of the deep empowerment of nursing research by artificial intelligence technology, traditional research ethics faces a double challenge: the ambiguity of the responsible party and the failure of ethical boundaries. When AI participates in, or even dominates, research areas such as data analysis and risk prediction, the responsible party becomes unclear. This ambiguity of responsibility makes it difficult to hold people accountable effectively and remedy problems when they occur. Current ethical norms are primarily designed for human behavior and are difficult to apply directly to AI systems that possess a degree of autonomy, can evolve dynamically, and employ opaque decision-making logic, creating a gap in ethical oversight in practice.^73,74 The emphasis on empathy and humanistic care in the nursing profession itself further exacerbates the complexity of ethical judgment. This dilemma of ambiguous responsibility has been confirmed in empirical research. The decision-making of artificial intelligence blurs the boundaries of responsibility between nurses, developers, and institutions, making it difficult to hold people accountable when adverse events occur. This conclusion is consistent with the research results of Wesam et al. ⁶¹ Vayena et al. ⁵⁷ also noted a general lack of accountability in the application of medical artificial intelligence. Nursing graduate students are more likely to fall into the trap of “not holding people accountable” and attribute the risks and responsibilities to the artificial intelligence developers. At present, there is no clear international division of responsibility for artificial intelligence-assisted research, and there is also a lack of established regulations for assigning responsibility in nursing research ethics education. The lack of responsibility determination and norms in the use of artificial intelligence by schools, hospitals and other related institutions has further led to the rationalization of students’ “lack of responsibility awareness”. At the same time, the blurring of ethical boundaries has become another urgent problem. Nursing graduate students generally report a lack of clarity about the compliance boundaries for the use of artificial intelligence. This phenomenon is consistent with the research results of Maggie et al., ⁷⁵ reflecting the reality that the development of artificial intelligence ethics norms lags behind technological development. Nursing research is both academic and practical, and its ethical boundaries cover multiple dimensions, including research integrity, patient rights protection and technology application. ⁷⁶ The American Nurses Association clearly stated in its position statement that nurses must ensure that advanced technology does not harm the core interpersonal and relational nature of the nursing profession. ⁷⁷ Elendu et al. ⁷⁸ also emphasized that traditional ethical principles such as autonomy, benevolence, non-harm and justice should continue to serve as the basis for guiding medical artificial intelligence practice. However, existing international ethical norms do not provide specific regulations for the application of artificial intelligence in nursing research, leaving practice without clear guidance. The emphasis on empathy and humanistic care in the nursing profession itself further exacerbates the complexity of ethical judgment. ⁷⁹

In response to shortcomings in both responsibility and ethics, the nursing profession needs to proactively develop ethical management capabilities suited to the intelligent era across both teaching and practical work. In education, artificial intelligence and nursing ethics should be included as core courses for postgraduate nursing students. Through scenario simulation, case studies and interdisciplinary dialogue, students’ ethical awareness and sense of responsibility should be cultivated. ⁸⁰ Studies have shown that structured artificial intelligence ethics education can effectively improve students’ ethical capabilities. ⁸¹ For example, the curriculum framework proposed by Emily et al. ⁸² integrates critical artificial intelligence literacy into pre-nursing education, aiming to systematically cultivate graduates’ core competencies in technology assessment, ethical awareness and clinical integration. Tuba et al. ⁸³ combined artificial intelligence with nursing ethical decision-making using the REST model, providing a theoretical framework for constructing an educational framework. In terms of practice and governance, a management system with clear responsibilities, traceability and supervision should be established to allocate responsibilities and conduct ethical reviews throughout the entire AI process, from research and development to use.^38,60 The core values of the nursing profession should always be upheld, the auxiliary role of artificial intelligence should be clearly defined, and the bottom line of humanistic care and professional ethics should be maintained. ⁷³ In terms of standard-setting, we can refer to the “cross-domain artificial intelligence principles” proposed in the UK government white paper and take nursing ethics theory as the core guide.^84,85 We should follow a multi-stakeholder collaboration model, with universities, medical institutions, technology developers and ethics experts jointly developing “Guidelines for the Ethical Application of Artificial Intelligence” specifically for nursing research, to clarify the responsibilities and behavioral boundaries of all parties. ⁸⁶ Only through the dual paths of educational improvement and institutional norms can we ensure that nursing research consistently adheres to rigorous ethical standards and professional responsibilities in the era of artificial intelligence, while promoting the empowerment of technology.

The vicious cycle of lack of support systems and inadequate coping capabilities

This study found that nursing graduate students generally face multiple challenges when dealing with the risks of artificial intelligence data, including weak risk awareness, insufficient protective measures and insufficient systematic training. This phenomenon reflects the lack of initiative in risk prevention and control among this group. This passive state is not only due to individual negligence, but also to the structural dilemma caused by the lack of a support system and individual coping ability. The failure of educational and clinical institutions to provide systematic data risk training, clear protection standards and related technical support has led to insufficient risk awareness and a lack of coping methods among nursing students and practitioners. Their weak ability further reduces their willingness to seek help actively or to promote improvements in the system, forming a vicious cycle. In the end, it not only threatens patient data security and nursing quality but also undermines the ethical autonomy and industry trust of the nursing profession in the digital age. Heinrichs et al. ⁸⁷ also noted that a lack of early technical experience and professional empowerment will continue to fuel practitioners’ concerns about artificial intelligence; by contrast, as familiarity with use and professional ability improve, such concerns will be significantly reduced. Existing practices have sought to improve clinical trainees’ understanding of AI principles and limitations through structured teaching, demonstrating the effectiveness of the educational intervention. ⁸⁸

To address this dilemma, the nursing education system needs to incorporate data literacy into the core competencies of nursing, offer compulsory courses in nursing informatics and data ethics at the postgraduate level and cultivate risk identification, ethical decision-making and cross-team communication skills through scenario simulation, case studies and interdisciplinary collaboration. ⁸⁹ Medical institutions should formulate clear guidelines for the use of AI tools and for data security operations, establish dedicated nursing information positions to facilitate technical communication, incorporate data risk management into quality evaluation, build a non-accountability, learning-oriented safety culture, and establish a feedback channel for school-hospital collaboration. Through collaborative development of dynamically updated training modules, establishment of clinical data security practice archives, and regular multidisciplinary seminars, the educational content and clinical risk scenarios can be updated synchronously to optimize the protection system continuously.

Legal constraints and platform protection are also key factors affecting the effectiveness of risk management. Minssen et al. ⁵⁷ noted that there is a general lack of laws and regulations worldwide that comprehensively regulate the use of data in artificial intelligence research. At present, most artificial intelligence research platforms lack data security features, such as data destruction after a response is completed. This reflects the disconnect between the supply of platform technical services and the data security needs of nursing research, which further exacerbates the prevention and control of risks. David et al. ⁹⁰ suggested that governments and universities should strengthen cooperation to jointly regulate the application of artificial intelligence technology in medical education. The nursing graduate students in this study called for stronger legal constraints, consistent with the international trend toward stronger regulation of artificial intelligence.

The specific demands raised by nursing graduate students in the interviews, such as the establishment of specialized courses and the formation of multidisciplinary consultation teams, provide a clear direction for building an effective support system. This is highly consistent with the recommendations of many institutions in the study by Paulina et al., ⁹¹ which emphasized “interdisciplinary cooperation” and “providing advanced artificial intelligence literacy training for graduate students.” The Nursing and Artificial Intelligence Leadership Collaboration Alliance also emphasizes that universities should establish interdisciplinary cooperation projects, such as joint training programs for nursing and technical students, to encourage students to directly participate in the development and governance of artificial intelligence technologies. ² This study suggests that universities can explore implementing a dual-mentor system for artificial intelligence research, with joint guidance from experts in nursing and information science. This will significantly enhance nursing graduate students’ ability to identify, assess, and respond to risks associated with artificial intelligence data, laying a solid foundation for them to conduct standardized, safe, and innovative intelligent-assisted research.

Regarding the relevance of this research and its topic to nursing and healthcare management, several key points can be discussed. This study, through interviews and qualitative topic analysis, identifies a five-dimensional structure of nursing graduate students’ perceptions of data risks in AI-enabled nursing research: data adaptation, data security, data quality, data ethics, and risk management. It highlights a comprehensive risk focus of “use-generation-sharing-responsibility,” offering empirical evidence for nursing managers to identify and address AI data risks. When promoting AI implementation, healthcare management should simultaneously build a data governance and quality assurance system, strengthen ethical compliance requirements, clarify responsibility boundaries, and institutionalize these practices across relevant research and management processes. Strengthening support for talent development and institutional supply is crucial to reducing systemic risk exposure. From a global perspective, the lag in current regulations and the inadequacy of platform security are prevalent in various contexts. Therefore, the findings of this study can provide empirical evidence for countries and institutions to build governance systems, talent development mechanisms, and standards for AI-assisted nursing research, ensuring synergistic optimization of safety and sustainability while improving efficiency (Figure 1).OPEN IN VIEWER