Moral Disengagement and Unethical Generative AI Use as the Chain Mediators Between Antagonistic Personality and Problematic Generative AI Use

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This is a highly original, well-structured article that is essential for advancing our understanding of generative artificial intelligence. It would be interesting to consider conducting a longitudinal study to track the variables over time and take samples from different cultures.

Author Response

RESPONSES TO REVIEWER 1

 

Comment 1: This is a highly original, well-structured article that is essential for advancing our understanding of generative artificial intelligence. It would be interesting to consider conducting a longitudinal study to track the variables over time and take samples from different cultures.

Response 1: Thank you for your positive evaluation and valuable suggestion. We agree that longitudinal and cross-cultural research would provide deeper insights into the dynamics of problematic generative AI use. This recommendation has already been acknowledged in the limitations section of the manuscript, where we highlight the need for future longitudinal and cross-cultural studies.

Reviewer 2 Report

Comments and Suggestions for Authors

The author deals with a very important topic. The theoretical background is thoroughly developed, and the methodology is also very impressive.

At the same time, the age characteristics of the sample (19–79-years-old) raise the possibility of using multiple sub-samples, which the author did not do. This would be professionally justified, as the IT knowledge and attitudes of a 19-year-old and a 79-year-old can be very different. It focuses on gender, but age could also be a factor. I suggest breaking down the sample and comparing these groups. Cultural, social, and professional characteristics will definitely play a role here, and these should also be taken into account. 

The discussion should be reconsidered and revised in light of this.

It is worth paying attention to the data in the tables; do not break a value or abbreviation in a row.

Once again, congratulations on your study. I am confident that, following the modification, the literature will be enriched with a study that is even more valuable from a developmental psychology perspective.

Comments on the Quality of English Language

The English could be improved to more clearly express the research.

Author Response

RESPONSES TO REVIEWER 2

 

Comment 1: At the same time, the age characteristics of the sample (19–79-years-old) raise the possibility of using multiple sub-samples, which the author did not do. This would be professionally justified, as the IT knowledge and attitudes of a 19-year-old and a 79-year-old can be very different. It focuses on gender, but age could also be a factor. I suggest breaking down the sample and comparing these groups. Cultural, social, and professional characteristics will definitely play a role here, and these should also be taken into account. 

The discussion should be reconsidered and revised in light of this.

Response 1: Thank you for this valuable comment. We would like to clarify that the age range of 19–79 years refers to the exploratory factor analysis (EFA) sample used during the scale development phase. The primary analyses reported in the manuscript were conducted on a different sample with an age range of 28–83 years. In addition, following the reviewer’s suggestion, we conducted additional analyses by dividing the main sample into age groups and examining the model separately across these groups. These results have been briefly reported in the revised manuscript.

 

Below parts were added to revised manuscript:

Abstract:

“A comparable pattern was also observed across age groups, with only minor variations in the mediation pathways.”

Methods section:

“To explore potential age-related differences, participants were categorized into two age groups (28–42 years and 43–83 years). Independent samples t-tests were conducted to examine group differences in the study variables. In addition, the proposed mediation model was tested separately for both age groups to explore potential differences in the pattern of relationships.”

Results section:

“To further explore potential age-related differences, the proposed mediation model was tested separately for two age groups (28–42 years and 43–83 years). Independent samples t-tests showed that the younger group reported significantly higher PGAIU scores than the older group (t[489] = 4.84, p < .01), while no significant differences were observed for the other variables.”

“When the mediation model was examined separately for each age group (not depicted in the figure), the overall pattern of relationships remained largely consistent across groups. The only notable difference concerned the association between moral disengagement and PGAIU: in the younger group, this relationship became fully indirect through the mediators (β= .14, p<.001; 95% CI [.06, .24]), whereas in the older group the partial mediation pattern observed in the total sample remained (β= .09, p<.01; 95% CI [.03, .17]). In addition, age-specific differences emerged in the pathway linking Machiavellianism to UGAIU. In the younger group, this association was partially mediated by moral disengagement (β= .11, p<.01; 95% CI [.05, .18]), whereas in the older group the relationship between Machiavellianism and UGAIU was fully mediated by moral disengagement (β= .09, p<.01; 95% CI [.03, .17]).”

Discussion section:

“Additional exploratory analyses considering age groups revealed broadly similar patterns across age categories. Although younger participants reported higher levels of PGAIU, the overall mediation structure remained largely consistent. However, the relationship between moral disengagement and PGAIU was fully indirect in the younger group, whereas a partial mediation pattern was observed among older participants, suggesting that age may slightly shape the pathways linking moral disengagement to PGAIU. One possible explanation is that younger individuals may be more likely to translate morally disengaged cognitions into problematic GAI engagement primarily through specific unethical use practices (Song & Liu, 2025). In other words, for younger users, moral disengagement may first facilitate the justification of unethical GAI use behaviors, which subsequently increases the likelihood of problematic engagement with generative AI tools (Karakuş et al., 2025). In contrast, older individuals may engage in PGAIU through additional pathways beyond unethical use practices, resulting in a partially mediated relationship.”

“Interestingly, age-specific differences also emerged in the Machiavellianism–UGAIU pathway, such that the association was fully mediated by moral disengagement among older participants but only partially mediated among younger users. This pattern may suggest that unethical GAI use among older individuals is more strongly dependent on cognitive moral justification processes, whereas younger users may engage in such behaviors through additional motivations (e.g., experimentation, instrumental convenience, or normative flexibility in digital environments).”

 

Comment 2: It is worth paying attention to the data in the tables; do not break a value or abbreviation in a row.

Response 2: Thank you for this helpful comment. The issue appears to have resulted from the journal’s formatting during the review process. Nevertheless, we carefully checked the tables in the revised manuscript to ensure that values and abbreviations are presented clearly without being split across rows.

Reviewer 3 Report

Comments and Suggestions for Authors

The manuscript addresses a genuine gap in the literature. While a small number of studies have examined Big Five correlates of PGAIU, no prior empirical work appears to have simultaneously investigated all three dark triad traits in conjunction with moral disengagement and UGAIU in an adult, primarily Western sample. The theoretical integration of the I-PACE model with social cognitive theory's account of moral disengagement is logical and well-articulated.

The development of the Unethical Generative AI Use Scale (UGAIUS) within the study is a notable contribution. The authors provide EFA and CFA results, report multiple reliability coefficients (Cronbach's alpha, McDonald's omega, composite reliability, and AVE), and describe item reduction procedures transparently. The factor structure is theoretically coherent, distinguishing academic integrity concerns, instrumental use orientation, and moral disengagement in AI use.

The analytic approach is appropriate. Bias-corrected bootstrapping with 10,000 resamples for indirect effects is consistent with contemporary standards, and the inclusion of daily GAI use frequency, age, and sex as covariates reflects reasonable methodological care. The multi-group comparison adds interpretive value, particularly regarding the differential role of moral disengagement among men and women.

The writing is clear and generally well-organized (though please see my caveat). The hypotheses follow logically from the literature review, and the discussion addresses both confirmatory and surprising findings.

I would recommend some slight revisions, based on the following:

1. Causal language throughout the manuscript. The study is cross-sectional and self-report-based throughout. Despite appropriate caveats in the Limitations section, causal or quasi-causal language appears recurrently in the Results and Discussion sections (e.g., "moral disengagement fosters," "psychopathic tendencies lead to," "narcissism drives"). Authors should systematically replace such language with explicitly associational phrasing. This is not a minor stylistic concern; it affects the validity of the interpretive claims. 
2. Common method bias. This is perhaps my major concern.  While the Harman single-factor test is reported (27.81% variance for one factor), this procedure is widely regarded as inadequate and even misleading as the sole test of common method variance. The authors are encouraged to acknowledge this limitation more explicitly and, where possible, reference additional strategies that could be employed in future work, such as the use of marker variables or confirmatory approaches to assessing method factors. If they are happy with this method, an appropriate and current citation in contemporary literature might be warranted.
3. UGAIUS as a newly developed scale. Although the scale development process is reported, the validation evidence is contained almost entirely within the same dataset used to test the primary hypotheses. This creates circularity in the findings: UGAIU's association with PGAIU could partly reflect shared method variance or shared item content, particularly given that several UGAIUS items reference cognitive justification of AI use in ways that overlap conceptually with the moral disengagement items. The authors should address this overlap directly. A brief discussion of how the UGAIUS discriminant validity was established relative to the Propensity to Morally Disengage Scale would be valuable. 
4. Directionality of moral disengagement. The authors briefly acknowledge that prior literature has modeled moral disengagement as a consequence rather than an antecedent of problematic technology use, and they justify their directional choice by reference to the I-PACE model. This justification is reasonable but could be more thoroughly defended. The decision to position moral disengagement as a predictor is theory-driven and the correlational data cannot adjudicate it. A stronger statement of this limitation, and a reference to longitudinal designs as the appropriate remedy, would strengthen the manuscript's epistemological grounding  
5. Sex versus gender (minor point!) The manuscript uses "sex" and "gender" somewhat interchangeably. Given that the variable was operationalized as a binary self-report item, the authors should clarify exactly how it was assessed and use consistent terminology throughout. If the item used the word "gender," that wording should be reflected in the text. 
6. Sample description in the abstract. The abstract reports the total sample as N = 851 (the combined EFA and CFA phases), while the primary path analytic findings are based on N = 491. This may mislead readers about the effective sample size for the study's main conclusions. The abstract should make the distinction explicit, or at a minimum, note that the primary analyses used N = 491.

Comments on the Quality of English Language

My concerns are quite minor and I believe they can be addressed very quickly. The English is fine but the causal language does not seem appropriate  (# 1 above)

 

Author Response

RESPONSES TO REVIEWER 3

 

Comment 1: Causal language throughout the manuscript. The study is cross-sectional and self-report-based throughout. Despite appropriate caveats in the Limitations section, causal or quasi-causal language appears recurrently in the Results and Discussion sections (e.g., "moral disengagement fosters," "psychopathic tendencies lead to," "narcissism drives"). Authors should systematically replace such language with explicitly associational phrasing. This is not a minor stylistic concern; it affects the validity of the interpretive claims. 

Response 1: Thank you for this important comment. Accordingly, we carefully revised the manuscript to replace causal or quasi-causal expressions (e.g., “fosters,” “leads to,” “drives”) with explicitly associational phrasing (e.g., “is associated with,” “is linked to,” “is related to”) throughout the Results and Discussion sections. These revisions were made to ensure that the interpretation of the findings accurately reflects the non-causal design of the study.

 

Comment 2: Common method bias. This is perhaps my major concern.  While the Harman single-factor test is reported (27.81% variance for one factor), this procedure is widely regarded as inadequate and even misleading as the sole test of common method variance. The authors are encouraged to acknowledge this limitation more explicitly and, where possible, reference additional strategies that could be employed in future work, such as the use of marker variables or confirmatory approaches to assessing method factors. If they are happy with this method, an appropriate and current citation in contemporary literature might be warranted.

Response 2: Thank you for raising this important point regarding common method bias. In response to this comment, we have clarified this issue more explicitly in the revised manuscript and acknowledged the limitations of relying solely on this procedure. We also noted that future research could employ additional approaches, such as marker variables or confirmatory factor analytic methods that model a latent method factor, to provide a more rigorous assessment of common method bias.

This revision has been incorporated into the Limitations section.

 

Below parts were added to revised manuscript:

Results section:

“While relying solely on this procedure has recognized limitations (Podsakoff et al., 2003), recent contemporary literature demonstrates that Harman’s test remains a valid and effective indicator when a study is built upon strong theorizing (Kock, 2021; Kock & Dow, 2025). Therefore, given our well-developed theoretical model, this result provides preliminary support that a single factor does not account for the majority of the variance. A more detailed discussion on the limitations of this approach and recommendations for future research are provided in the Limitations section.”

Limitations section:

“Another limitation concerns the potential influence of common method variance, as all variables were measured using self-report instruments within a single survey. Although Harman’s single-factor test suggested that a single factor did not account for the majority of variance, this procedure has recognized limitations and should not be considered a definitive test of common method bias. Future studies could employ additional strategies, such as marker variables or confirmatory factor analytic approaches that model a latent method factor, to more rigorously assess potential method effects.”

 

Comment 3: UGAIUS as a newly developed scale. Although the scale development process is reported, the validation evidence is contained almost entirely within the same dataset used to test the primary hypotheses. This creates circularity in the findings: UGAIU's association with PGAIU could partly reflect shared method variance or shared item content, particularly given that several UGAIUS items reference cognitive justification of AI use in ways that overlap conceptually with the moral disengagement items. The authors should address this overlap directly. A brief discussion of how the UGAIUS discriminant validity was established relative to the Propensity to Morally Disengage Scale would be valuable. 

Response 3: Thank you for this thoughtful comment. We agree that conceptual overlap between constructs should be carefully considered. However, the correlation between the Propensity to Morally Disengage Scale and the UGAIUS moral disengagement–related dimension was moderate (r = .41), which suggests that the constructs are related but empirically distinct. This magnitude of association is consistent with theoretical expectations, as moral disengagement represents a general cognitive mechanism for justifying unethical behavior, whereas UGAIUS captures context-specific tendencies related to unethical uses of generative AI systems. We have clarified this distinction in the revised manuscript.

 

Below parts were added to revised manuscript:

 

Discussion section:

“Importantly, the moderate association between moral disengagement and UGAIUS observed in the present study supports the conceptual distinction between a general propensity to morally disengage and context-specific unethical AI use behaviors. In addition, the theoretically consistent associations observed between UGAIUS and related constructs provide preliminary support for the construct validity of the scale.”

 

Limitations section:

“Although the Unethical Generative AI Use Scale (UGAIUS) and the Propensity to Morally Disengage Scale are theoretically related, they capture conceptually distinct constructs. Moral disengagement reflects generalized cognitive mechanisms that enable individuals to justify unethical behavior across a wide range of contexts, whereas UGAIUS focuses specifically on behavioral tendencies related to the unethical use of generative AI systems. Nevertheless, because the scale validation and hypothesis testing were conducted within the same dataset, some degree of shared method variance or conceptual overlap cannot be entirely ruled out. Future research would benefit from further validating the UGAIUS using independent samples and additional methodological approaches to strengthen evidence for its discriminant validity.”

 

Comment 4: Directionality of moral disengagement. The authors briefly acknowledge that prior literature has modeled moral disengagement as a consequence rather than an antecedent of problematic technology use, and they justify their directional choice by reference to the I-PACE model. This justification is reasonable but could be more thoroughly defended. The decision to position moral disengagement as a predictor is theory-driven and the correlational data cannot adjudicate it. A stronger statement of this limitation, and a reference to longitudinal designs as the appropriate remedy, would strengthen the manuscript's epistemological grounding.  

Response 4: We thank the reviewer for this observation. In line with the suggestion, we have strengthened the introduction and limitations sections to more explicitly acknowledge the directional ambiguity of the moral disengagement–PGAIU relationship and to recommend longitudinal designs as the appropriate methodological remedy.

 

Below parts were added to revised manuscript:

Introduction section:

“Once activated, moral disengagement creates the cognitive conditions necessary for unethical behavior to occur — and, we argue, for unethical GAI use specifically. In digital environments, GAI technologies introduce a form of psychological distance between the user and the ethical implications of their actions: the tool performs the work, the output appears legitimate, and accountability is diffused across human and algorithmic agents (Shaw, 2025; Sun et al., 2025). This architecture maps directly onto Bandura's (2002) mechanisms of moral disengagement, particularly displacement of responsibility and distortion of consequences, rendering GAI a uniquely permissive context for morally disengaged individuals. Consequently, we expect moral disengagement to mediate the association between antagonistic traits and UGAIU.

         Beyond its role as a cognitive enabler of unethical behavior, moral disengagement may also directly elevate the risk of PGAIU through a distinct pathway. Prior research has documented associations between moral disengagement and various forms of problematic technology use, including online gaming, social media, and generalized internet use (Colella et al., 2024; Kocabıyık, 2026; Xiao & Cheng, 2023). Importantly, much of this literature has examined moral disengagement as a consequence of problematic use — positing that excessive technology engagement gradually erodes ethical inhibitions. While this directionality is plausible, the predominantly cross-sectional and correlational nature of existing studies precludes definitive causal inference, and the reverse pathway remains equally theoretically defensible. Drawing on the I-PACE model (Brand et al., 2019), we position moral disengagement as a predispositional cognitive response that operates upstream of problematic use: individuals who habitually disengage their moral standards are less likely to exercise the self-regulatory restraint needed to moderate technology use, and more likely to exploit GAI's mood-modifying and effort-reducing properties as a maladaptive coping strategy (Goh et al., 2025). This is consistent with evidence that morally disengaged individuals show impaired self-regulation and heightened negative affect (Coşkun, 2025; Lu et al., 2025), both of which are established risk factors for behavioral addiction (Brand et al., 2019). The present study therefore contributes to this literature by positioning moral disengagement as a theoretically grounded antecedent — rather than consequence — of PGAIU, and by embedding this directional claim within an integrative personality-based model.”

Limitations section:

“a notable limitation concerns the directionality of the moral disengagement–PGAIU relationship. Although the present model positions moral disengagement as a theoretically grounded antecedent of PGAIU, consistent with the I-PACE framework, the cross-sectional design does not permit causal inference. The reverse pathway — whereby problematic GAI use gradually erodes moral standards — is equally plausible and cannot be ruled out on the basis of the present data. Future longitudinal and experience-sampling studies tracking moral disengagement and PGAIU across multiple time points would be essential to empirically adjudicate this directionality.”

 

Comment 5: Sex versus gender (minor point!) The manuscript uses "sex" and "gender" somewhat interchangeably. Given that the variable was operationalized as a binary self-report item, the authors should clarify exactly how it was assessed and use consistent terminology throughout. If the item used the word "gender," that wording should be reflected in the text. 

Response 5: Thank you for this helpful observation. The manuscript has been revised to ensure consistent use of the term “gender” throughout. The remaining instance of “sex” has been corrected to reflect the wording used in the survey item.

 

Comment 6: Sample description in the abstract. The abstract reports the total sample as N = 851 (the combined EFA and CFA phases), while the primary path analytic findings are based on N = 491. This may mislead readers about the effective sample size for the study's main conclusions. The abstract should make the distinction explicit, or at a minimum, note that the primary analyses used N = 491.

Response 6: Thank you for this helpful comment. To avoid potential confusion regarding the effective sample size used for the primary analyses, the abstract has been revised and the reported sample size has been corrected to N = 491.

Reviewer 4 Report

Comments and Suggestions for Authors

The manuscript examines the relationships between antagonistic personality traits (narcissism, Machiavellianism, and psychopathy) and problematic generative AI use (PGAIU), proposing moral disengagement and unethical generative AI use (UGAIU) as sequential mediators. Using survey data from adult participants and structural path modeling, the study attempts to explain psychological mechanisms underlying problematic engagement with generative AI technologies. The topic is timely and relevant given the rapid adoption of generative AI tools in academic and professional contexts.

Overall, the manuscript presents an interesting theoretical integration of personality psychology, moral cognition, and technology use. The empirical analysis is generally well-structured, and the statistical approach is appropriate for the research questions. However, several conceptual, methodological, and presentation issues should be addressed before the manuscript can be considered for publication.

Strengths

1. Timely and relevant topic – The study addresses an emerging phenomenon: problematic use of generative AI tools and their ethical implications.
   2. Clear theoretical framework– The study uses the I-PACE model to structure the hypotheses and explain how personality traits may lead to problematic AI use.
   3. Large sample size– The sample size (N = 851 across two phases) is adequate for factor analysis and path modeling.
   4. Use of advanced statistical modeling – The combination of exploratory factor analysis (EFA), confirmatory factor analysis (CFA), and path analysis strengthens the methodological rigor.
   5. Development of a new scale – The Unethical Generative AI Use Scale (UGAIUS) represents a potentially valuable contribution to the literature if further validated.

Concerns

1. Conceptual clarity of “problematic AI use”

The manuscript refers to problematic generative AI use as a form of technology addiction, yet the theoretical distinction between problematic use, dependency, and addiction remains unclear. A clearer conceptual definition and differentiation from general problematic technology use is needed.

2. Cross‑sectional design and causal interpretation**

Although the model implies directional pathways (e.g., personality → moral disengagement → unethical AI use → problematic AI use), the data are cross‑sectional. The manuscript occasionally interprets results in causal language, which should be toned down.

3. Measurement validity of the new scale**

The newly developed UGAIUS scale is promising but requires stronger validation. The manuscript should provide additional evidence of construct validity, such as comparisons with related constructs or behavioral indicators.

4. Cultural and contextual limitations

Participants were primarily recruited via Prolific and appear to represent English-speaking Western users. This limits the generalizability of the findings, particularly given that AI use practices may vary substantially across cultural and educational contexts.

5. Conceptual overlap between constructs

Some dimensions of unethical AI use appear conceptually close to moral disengagement, which may inflate the mediation effect. The manuscript should discuss the potential overlap and explain how the constructs are empirically distinct.

Author Response

RESPONSES TO REVIEWER 4

 

Comment 1: Conceptual clarity of “problematic AI use”

The manuscript refers to problematic generative AI use as a form of technology addiction, yet the theoretical distinction between problematic use, dependency, and addiction remains unclear. A clearer conceptual definition and differentiation from general problematic technology use is needed.

Response 1: We thank the reviewer for highlighting the need for conceptual clarity regarding 'problematic AI use'. In the revised manuscript, we have expanded the paragraph to explicitly define these terms. Specifically, we conceptualized 'problematic use' as an umbrella term that captures dysregulated engagement leading to negative life outcomes, distinguishing it from the strict clinical criteria of 'addiction' and the specific reliance emphasized in 'dependency'. Furthermore, we clarified that generative AI differs from general problematic technology (like internet or social media) because it acts as an active 'cognitive partner' rather than just a passive medium, uniquely fostering cognitive miserliness and pseudosocial bonds. The updated paragraph now reflects these theoretical boundaries comprehensively.

 

Below parts were added to revised manuscript:

“While often used interchangeably, these concepts possess distinct theoretical nuances. Addiction typically denotes severe, compulsive behavioral patterns accompanied by tolerance and withdrawal-like symptoms (Yankouskaya et al., 2025; Yu et al., 2024). Dependency highlights a deep functional or socio-emotional reliance where individuals feel incapable of functioning without the tool (Yankouskaya et al., 2025; Zhang et al., 2024). Conversely, problematic use serves as a broader umbrella term encompassing excessive and dysregulated engagement that impairs daily functioning, without necessarily meeting strict clinical diagnostic criteria for addiction (Yu et al., 2024). Furthermore, problematic GAI use differs fundamentally from general problematic technology use (e.g., internet or social media addiction). While traditional digital technologies often involve passive content consumption or act merely as communication channels between humans, generative AI functions as an active 'cognitive partner' (Yankouskaya et al., 2025). It provides highly personalized, context-aware interactions that not only foster cognitive miserliness by encouraging users to offload mental effort (Deng & Deng, 2025), but also simulate human-like empathy, leading to unique pseudosocial or parasocial bonds (Hu et al., 2023; Huang et al., 2025; Yankouskaya et al., 2025). To reduce conceptual confusion and capture this broader spectrum of dysregulated human-GAI interaction, the present study adopts the term problematic generative artificial intelligence use (PGAIU), which refers to difficulties in controlling GAI use and the experience of negative outcomes in daily life (Goh et al., 2025).”

 

Comment 2: Cross‑sectional design and causal interpretation**

Although the model implies directional pathways (e.g., personality → moral disengagement → unethical AI use → problematic AI use), the data are cross‑sectional. The manuscript occasionally interprets results in causal language, which should be toned down.

Response 2: Thank you for this important comment. We agree that causal language should be avoided given the cross-sectional design of the study. The manuscript has been carefully revised to replace causal expressions with associational wording throughout the Results and Discussion sections. Please see the detailed response to Reviewer 3 for further clarification.

 

 

Comment 3: Measurement validity of the new scale**

The newly developed UGAIUS scale is promising but requires stronger validation. The manuscript should provide additional evidence of construct validity, such as comparisons with related constructs or behavioral indicators.

Response 3: Thank you for this constructive suggestion. In the manuscript, the development and validation of the UGAIUS included several standard psychometric procedures, including factor analytic evidence supporting its dimensional structure as well as satisfactory reliability estimates. In addition, the scale demonstrated theoretically meaningful associations with related constructs, such as problematic generative AI use and moral disengagement, which provide preliminary evidence of construct validity. Nevertheless, we acknowledge that further validation would strengthen confidence in the scale. Future research could examine the scale’s associations with additional related constructs and behavioral indicators of unethical AI use, as well as replicate the findings in independent samples. Please also see the detailed response to Reviewer 3 for further clarification regarding the UGAIUS scale.

 

Below parts were added to revised manuscript:

“Importantly, the moderate association between moral disengagement and UGAIUS observed in the present study supports the conceptual distinction between a general propensity to morally disengage and context-specific unethical AI use behaviors. In addition, the theoretically consistent associations observed between UGAIUS and related constructs provide preliminary support for the construct validity of the scale.”

 

 

Comment 4: Cultural and contextual limitations

Participants were primarily recruited via Prolific and appear to represent English-speaking Western users. This limits the generalizability of the findings, particularly given that AI use practices may vary substantially across cultural and educational contexts.

Response 4: Thank you for this important observation. We agree that cultural and contextual factors may influence AI use practices. This limitation has already been acknowledged in the manuscript, where we note that the sample primarily consisted of English-speaking Western users recruited via Prolific, which may limit the generalizability of the findings across different cultural and educational contexts.

 

Comment 5: Conceptual overlap between constructs

Some dimensions of unethical AI use appear conceptually close to moral disengagement, which may inflate the mediation effect. The manuscript should discuss the potential overlap and explain how the constructs are empirically distinct.

Response 5: Thank you for this thoughtful comment. We agree that potential conceptual overlap between constructs should be carefully considered. However, the correlation between the Propensity to Morally Disengage Scale and the UGAIUS moral disengagement–related dimension was moderate (r = .41), suggesting that the constructs are related but empirically distinct. This magnitude of association is consistent with theoretical expectations, as moral disengagement represents a general cognitive mechanism that facilitates the justification of unethical behavior, whereas UGAIUS captures context-specific tendencies related to unethical uses of generative AI systems. Therefore, while the constructs are theoretically linked, the observed correlation does not indicate substantial redundancy that would artificially inflate the mediation effects. We have clarified this conceptual distinction in the revised Discussion section.

 

Below parts were added to revised manuscript:

 

Discussion section:

“Importantly, the moderate association between moral disengagement and UGAIUS observed in the present study supports the conceptual distinction between a general propensity to morally disengage and context-specific unethical AI use behaviors. In addition, the theoretically consistent associations observed between UGAIUS and related constructs provide preliminary support for the construct validity of the scale.”