Convenience Amid Crisis: Explaining Generative AI Misuse and Compliance Intention Among Myanmar University Students

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The Theory of Planned Behavior was extended to dishonest behavior early on, with work showing it could predict cheating, lying, and shoplifting from the same three components.

“TPB was extended to dishonest behavior specifically almost as soon as it was formulated, with early work showing it could predict cheating, lying, and shoplifting from the same three components (Beck & Ajzen, 1991;Dewanti et al., 2020;Harding et al., 2007).”

• cites Beck & Ajzen, 1991 sourceNot checked — no full text of the source
• cites Dewanti et al., 2020 sourcePartly supported

  The Introduction (page unavailable) identifies TPB’s three core constructs and applies the theory to predicting students’ cheating intentions. It does not say that early work predicted lying or shoplifting, or establish how soon after TPB’s formulation that work appeared. It also discusses moral obligation and past behavior alongside the three constructs. (LLM, conf=0.86)

  Experimental — accuracy not yet measured
• cites Harding et al., 2007 sourceNot checked — no full text of the source

Classified as probably not a substantial claim.

The Theory of Planned Behavior's central claim is that intended action follows from how a person feels about the act itself, what they believe others expect of them, and how much control they feel they have over doing it.

“Its central claim is simple: intended action follows from how a person feels about the act itself, what they believe others expect of them, and how much control they feel they have over doing it.”

• cites Ajzen (1991) sourceSupported

May combine two claims.

AI content detectors produce inconsistent results and are readily circumvented by minor edits to AI-generated text, limiting the weight detection-based enforcement alone can carry.

“AI content detectors have been shown to produce inconsistent results and are readily circumvented by minor edits to AI-generated text, which limits how much weight detection-based enforcement alone can carry (Hadra et al., 2026;Van Vlasselaer et al., 2026).”

• cites Hadra et al., 2026 sourcePartly supported

  The ‘Reconsidering the role of detection in academic integrity’ section reports inconsistent performance and says detectors should not be primary or determinative evidence in misconduct investigations. The Introduction reports that paraphrasing sharply reduced GPT-Zero’s accuracy, but the paper does not establish that AI detectors generally are readily circumvented by minor edits. Its ‘Limitations and future research’ section says the effects of editing AI-generated text to varying degrees were beyond this study’s scope. (LLM, conf=0.86)

  Experimental — accuracy not yet measured
• cites Van Vlasselaer et al., 2026 sourcePartly supported

  The Classification accuracy and error analysis section reports sharply different results across detectors, including reduced accuracy for Turnitin after AI-generated passages were humanised. The Discussion says detector flags should prompt further review rather than serve as a final conclusion. But the tested alteration was a GPT-4o humanising prompt, not necessarily a minor edit, and Pangram remained largely unaffected; the paper does not establish that detectors are generally readily circumvented by minor edits. Page numbers were unavailable in our copy. (LLM, conf=0.87)

  Experimental — accuracy not yet measured

Classified as probably not a substantial claim. May combine two claims.

Team-based and collaboration-oriented assessment formats have been proposed as a way to reduce opportunities for undisclosed AI reliance while building independent academic skill.

“Team-based and collaboration-oriented assessment formats, for instance, have been proposed specifically as a way to reduce opportunities for undisclosed AI reliance while also building the kind of independent academic skill that this study's Behavioral Control findings suggest matters (Issa & Hall, 2024).”

• cites Issa & Hall, 2024 sourceSupported

May combine two claims.

Academic Pressure positively predicted AI Misuse and negatively predicted Compliance Intention.

“Academic Pressure played a dual role, raising AI Misuse while lowering Compliance Intention”

• Also stated in: § 4.6. Regression predicting compliance intention · plain

  “Academic Pressure's negative association with Compliance Intention striking precisely because the same variable positively predicted AI Misuse.”

• AssumptionRests on an assumption: Self-reported AI Misuse accurately reflects students' actual unauthorized AI use.

May combine two claims.

AI Familiarity did not significantly predict either AI Misuse or Compliance Intention.

“AI Familiarity did not significantly predict either outcome”

• Also stated in: § 4.4. Bivariate correlations · plain

  “As expected, AI Familiarity was not a significant predictor here once attitudinal and institutional variables were in the model, and neither was Subjective Norms.”

• AssumptionRests on an assumption: Self-reported AI Misuse accurately reflects students' actual unauthorized AI use.

May combine two claims.

Compliance Intention rose most with Behavioral Control, Institutional Enforcement, and Subjective Norms, and fell with Academic Pressure.

“Compliance Intention rose most with Behavioral Control, Institutional Enforcement, and Subjective Norms, and fell with Academic Pressure.”

May combine two claims.

AI Misuse rose most with Misuse Acceptability and Moral Rationalization, and fell most with Institutional Enforcement.

“AI Misuse rose most with Misuse Acceptability and Moral Rationalization, and fell most with Institutional Enforcement.”

• AssumptionRests on an assumption: Self-reported AI Misuse accurately reflects students' actual unauthorized AI use.

May combine two claims.

The Compliance Intention regression model was significant, R2 = .215; adjusted R2 = .206; F(10, 840) = 22.99, p < .001.

“Note.R² = .215;adjusted R² = .206;F(10, 840) = 22.99, p < .001.”

May combine two claims.

AI Familiarity, Learning Impact, and Subjective Norms were not significant predictors of AI Misuse.

“| AI Familiarity | 0.038 | 0.032 | 0.027 | 1.197 | .232 | [-0.021, 0.086] | 1.16 |”

• Also stated in: § Section M -Intentions (Next Semester) · plain

  “| Learning Impact | -0.017 | -0.014 | 0.027 | -0.538 | .591 | [-0.067, 0.038] | 1.10 |”

• Also stated in: § Section M -Intentions (Next Semester) · plain

  “| Subjective Norms | -0.053 | -0.054 | 0.031 | -1.766 | .078 | [-0.114, 0.006] | 1.07 |”

May combine two claims.

In the AI Misuse model, Academic Pressure and Productive AI Use both emerged as significant positive predictors.

“Academic Pressure and Productive AI Use both emerged as significant positive predictors.”

• AssumptionRests on an assumption: Self-reported AI Misuse accurately reflects students' actual unauthorized AI use.

May combine two claims.

Compliance Intention, Behavioral Control, and Subjective Norms all sat above the midpoint of the seven-point scale, while Rule Clarity, Institutional Enforcement, and Moral Rationalization clustered closer to the midpoint.

“Compliance Intention, Behavioral Control, and Subjective Norms all sat above the midpoint of the seven-point scale, while Rule Clarity, Institutional Enforcement, and Moral Rationalization clustered closer to the midpoint.”

May combine two claims.

Academic Pressure had a negative association with Compliance Intention, which the study's hypotheses did not anticipate.

“A more unexpected result, one the study's hypotheses did not anticipate, was Academic Pressure's negative association with Compliance Intention striking precisely because the same variable positively predicted AI Misuse.”

May combine two claims.

AI Misuse and Compliance Intention are driven by different sets of predictors, with Academic Pressure oppositely signed.

“Its most distinctive contribution is the finding that these two outcomes are driven by different sets of predictors oppositely signed ones, in the case of Academic Pressure.”

May combine two claims.

In the AI Misuse model, students who found the rules clearer and enforcement more consistent reported less misuse once other factors were held constant, and Behavioral Control remained a significant negative predictor.

“Students who found the rules clearer and enforcement more consistent reported less misuse once other factors were held constant, and Behavioral Control likewise remained a significant negative predictor.”

• AssumptionRests on an assumption: Self-reported AI Misuse accurately reflects students' actual unauthorized AI use.

May combine two claims.

AI Familiarity, Productive AI Use, Learning Impact, and Moral Rationalization were not significant predictors of Compliance Intention.

“| AI Familiarity | -0.030 | -0.029 | 0.031 | -0.918 | .359 | [-0.090, 0.032] | 1.16 |”

• Also stated in: § Section M -Intentions (Next Semester) · plain

  “| Productive AI Use | 0.011 | 0.012 | 0.037 | 0.333 | .739 | [-0.060, 0.084] | 1.19 |”

• Also stated in: § Section M -Intentions (Next Semester) · plain

  “| Learning Impact | 0.026 | 0.025 | 0.031 | 0.807 | .420 | [-0.035, 0.085] | 1.10 |”

May combine two claims.

Mean scores were highest for Academic Pressure, Productive AI Use, and AI Familiarity, and lowest for AI Misuse and Misuse Acceptability.

“Mean scores were highest for Academic Pressure, Productive AI Use, and AI Familiarity, and lowest for AI Misuse and Misuse Acceptability”

May combine two claims.

Subjective Norms was not a significant predictor in the AI Misuse model but was the third-strongest predictor of Compliance Intention.

“In the AI Misuse model, Subjective Norms was not a significant predictor once other variables were controlled.In the Compliance Intention model, it was the third-strongest predictor of the ten, trailing only Behavioral Control and Institutional Enforcement.”

• AssumptionRests on an assumption: Self-reported AI Misuse accurately reflects students' actual unauthorized AI use.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Women made up the majority of the sample, first-year students were the largest single cohort by year of study, and STEM was the most common field represented.

“Women made up the majority of the sample, first-year students were the largest single cohort by year of study, and STEM was the most common field represented.”

May combine two claims.

In the Compliance Intention model, AI Familiarity and Productive AI Use were not significant predictors.

“As anticipated, AI Familiarity and Productive AI Use were not significant predictors in this model.”

May combine two claims. The quote only partly states this claim.

The AI Misuse regression model was significant, R2 = .284; adjusted R2 = .275; F(10, 840) = 33.30, p < .001.

“Note.R² = .284;adjusted R² = .275;F(10, 840) = 33.30,p < .001.”

• AssumptionRests on an assumption: Self-reported AI Misuse accurately reflects students' actual unauthorized AI use.

May combine two claims.

AI Misuse had the strongest reliability of the set, and Compliance Intention the weakest.

“AI Misuse had the strongest reliability of the set, and Compliance Intention the weakest”

May combine two claims.

Twelve eigenvalues exceeded 1.0, matching the number of hypothesized constructs exactly.

“Twelve eigenvalues exceeded 1.0, matching the number of hypothesized constructs exactly.”

May combine two claims.

Every scale met an acceptable reliability threshold and was carried forward into the regression analyses.

“every scale nonetheless met an acceptable threshold and was carried forward into the regression analyses.”

May combine two claims.

Sampling adequacy was excellent (Kaiser-Meyer-Olkin measure = .838) and Bartlett's test of sphericity was significant, χ2(946) = 10336.2, p < .001.

“Sampling adequacy was excellent (Kaiser-Meyer-Olkin measure = .838),and Bartlett's test of sphericity was significant, χ²(946) = 10336.2,p < .001”

Probably not a substantial claim: may not pass S3 (load-bearing).

Item F6 was excluded from the AI Misuse composite because it measures perceptions of peers rather than one's own conduct.

“One additional item in the AI Misuse block, F6, asked whether respondents knew classmates who had used AI improperly; since this measures perceptions of peers rather than one's own conduct, it was excluded from the AI Misuse composite.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Myanmar higher education has weathered substantial disruption since 2021, including extended closures, displacement of students and faculty, and a growing mix of alternative and online providers.

“Myanmar higher education has weathered substantial disruption since 2021, including extended closures, displacement of students and faculty, and a growing mix of alternative and online providers operating alongside traditional institutions (Htut et al., 2022;Nanthakorn et al., 2023).”

• cites Htut et al., 2022 sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

• cites Nanthakorn et al., 2023 sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

May combine two claims.

Research applying moral disengagement theory to ChatGPT-based plagiarism found that disengagement from one's own moral standards significantly predicted misuse, and that this effect grew stronger rather than weaker under formal sanctions.

“research applying moral disengagement theory to ChatGPT-based plagiarism found that disengagement from one's own moral standards, not just rational cost-benefit calculation, significantly predicted misuse, and that this effect grew stronger rather than weaker under formal sanctions.”

• cites Zhang et al., 2024Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Classified as probably not a substantial claim.

Cressey's fraud triangle argued that violation follows from the combination of felt pressure, a perceived opportunity, and an available rationalization.

“Cressey's (1953) fraud triangle argued that violation follows from the combination of felt pressure, a perceived opportunity, and an available rationalization.”

• cites Cressey (1953)Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Classified as probably not a substantial claim.

Disruption is treated throughout as interpretive context for the findings, not as a variable the analysis tests.

“Disruption is therefore treated throughout as interpretive context for the findings, not as a variable the analysis tests.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

An attention-check item was embedded to characterize response quality, and respondents who failed it were retained in the analytic sample rather than excluded.

“An attention-check item, instructing respondents to select a specific response option, was embedded within the instrument to characterize response quality; respondents who failed this item were retained in the analytic sample rather than excluded”

• AssumptionRests on an assumption: Retaining respondents who failed the attention-check item does not bias the regression results.

May combine two claims.

Crisis-era strain and thin institutional governance appear to be the more decisive forces.

“Crisis-era strain and thin institutional governance appear to be the more decisive forces, though this study's design can only speak to that connection as informed interpretation, not direct proof.”

Probably not a substantial claim: may not pass S3 (load-bearing).

A confirmatory factor analysis was not conducted.

“A confirmatory factor analysis, which tests a pre-specified measurement model directly rather than allowing structure to emerge from the data, was not conducted and remains a natural next step for future work seeking to confirm this measurement structure in an independent sample.”

Classified as probably not a substantial claim. May combine two claims.

The survey did not directly ask about institutional disruption history, mode of instruction, or displacement status.

“The survey also did not directly ask about institutional disruption history, mode of instruction, or displacement status.”

May combine two claims.

Productive AI Use and AI Familiarity are not the problem institutions need to manage.

“None of this makes Productive AI Use or AI Familiarity the problem institutions need to manage.”

May combine two claims.

Residuals did not depart significantly from normality for AI Misuse (Shapiro-Wilk W = .998, p = .648) and Compliance Intention (W = .998, p = .504).

“residuals did not depart significantly from normality (Shapiro-Wilk, AI Misuse: W = .998,p = .648;Compliance Intention: W = .998,p = .504)”

Probably not a substantial claim: may not pass S3 (load-bearing).

Checks of skewness and kurtosis revealed no meaningful departure from normality.

“Checks of skewness and kurtosis revealed no meaningful departure from normality.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Of the 851 participants, 839 (98.6%) passed the embedded attention check and 12 (1.4%) failed it.

“Among the 851 participants, 839 (98.6%) passed the embedded attention check, with 12 (1.4%)failing it.”

Probably not a substantial claim: may not pass S3 (load-bearing).

All measures relied on self-report, which carries the usual risk of social-desirability bias.

“All measures relied on self-report, which carries the usual risk of social-desirability bias.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The dataset contained no missing responses.

“The dataset contained no missing responses.”

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

Enforcement-only approaches may fall short where students are genuinely overloaded, and easing unnecessary pressure could reduce unauthorized use while strengthening stated compliance.

“Enforcement-only approaches may fall short where students are genuinely overloaded, and easing unnecessary pressure, or offering legitimate ways to manage it, could reduce unauthorized use while also strengthening stated compliance.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The survey collected no names, ID numbers, or other information that could identify a respondent.

“The survey collected no names, ID numbers, or other information that could identify a respondent.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Variance inflation factors were used to check multicollinearity among predictors, and the Breusch-Pagan test, the Ramsey RESET test, and the Shapiro-Wilk test on model residuals were used to check homoscedasticity, functional form, and residual normality, respectively.

“Variance inflation factors were used to check for multicollinearity among predictors, and the Breusch-Pagan test, the Ramsey RESET test, and the Shapiro-Wilk test on model residuals were used to check homoscedasticity, functional form, and residual normality, respectively.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The Breusch-Pagan test found no evidence of heteroscedasticity for AI Misuse (p = .091) and Compliance Intention (p = .543).

“the Breusch-Pagan test found no evidence of heteroscedasticity (AI Misuse: p = .091;Compliance Intention: p = .543)”

Probably not a substantial claim: may not pass S3 (load-bearing).

Institutional permission to administer the survey was obtained directly from a representative at each of the six participating universities prior to data collection.

“Institutional permission to administer the survey was obtained directly from a representative at each of the six participating universities prior to data collection.”

Classified as probably not a substantial claim.

A two-tailed .05 threshold governed judgments of statistical significance throughout.

“A two-tailed .05threshold governed judgments of statistical significance throughout.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

As an anonymous survey involving no intervention, the research qualified as minimal risk, and formal institutional ethics committee review was not required under the procedures applicable to this study context.

“As an anonymous survey involving no intervention, this research qualified as minimal risk, and formal institutional ethics committee review was not required under the procedures applicable to this study context.”

May combine two claims.

A longitudinal application of TPB to ChatGPT-specific cheating found that attitudes, Subjective Norms, and perceived Behavioral Control each predicted intentions to use chatbot-generated text dishonestly, and that those intentions predicted later self-reported use.

“a longitudinal application of TPB to ChatGPT-specific cheating found that attitudes, Subjective Norms, and perceived Behavioral Control each predicted intentions to use chatbot-generated text dishonestly, and that those intentions in turn predicted later self-reported use (Greitemeyer & Kastenmüller, 2024).”

• cites Greitemeyer & Kastenmüller, 2024 sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

The raw dataset was first checked for completeness, internal consistency, attention-check pass rates, and any evident entry errors.

“The raw dataset was first checked for completeness, internal consistency, attentioncheck pass rates, and any evident entry errors.”

Classified as probably not a substantial claim.

The absence of these factors means the disruption framing remains an interpretive lens rather than a tested explanation.

“Their absence means the disruption framing offered throughout this discussion remains an interpretive lens rather than a tested explanation.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The Ramsey RESET test found no evidence of functional-form misspecification for AI Misuse (p = .151) and Compliance Intention (p = .366).

“the Ramsey RESET test found no evidence of functional-form misspecification (AI Misuse: p = .151;Compliance Intention: p = .366)”

Classified as probably not a substantial claim. May combine two claims.

The survey measures attitudes, institutional perceptions, and behavior, not disruption exposure directly.

“But the survey itself measures attitudes, institutional perceptions, and behavior, not disruption exposure directly.”

May combine two claims.

The Subjective Norms composite mixes item types: three items (H1, H2, H4) tap injunctive norms, while reverse-scored H3 taps a descriptive norm.

“The Subjective Norms composite also mixes item types: three items (H1, H2, H4) tap injunctive norms, or perceived disapproval and its reputational cost, while the reverse-scored H3 taps a descriptive norm, students' beliefs about how common rule-violating AI use is among their peers.”

• AssumptionRests on an assumption: The Subjective Norms items H1, H2, H4 tap injunctive norms and H3 taps a descriptive norm.
• AssumptionRests on an assumption: Injunctive and descriptive norms are conceptually distinct constructs in the norms literature.

May combine two claims.

The more decisive levers are how acceptable unauthorized use feels, how clearly it is governed, and how much pressure students are under.

“The more decisive levers are how acceptable unauthorized use feels, how clearly it is governed, and how much pressure students are under when the temptation to cross that line arises.”

Probably not a substantial claim: may not pass S3 (load-bearing).

A decade-spanning review of academic cheating research found that contextual factors, especially students' perceptions of peers' behavior and their assessment of how likely detection is, predict dishonesty more powerfully than any individual trait.

“A decade-spanning review of academic cheating research reached a similar conclusion through a different route, finding that contextual factors, especially students' perceptions of peers' behavior and their assessment of how likely detection is, predict dishonesty more powerfully than any individual trait (McCabe et al., 2001).”

• cites McCabe et al., 2001 sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

In one of the largest studies of student attitudes toward generative AI to date, most students reported broadly positive views of these tools for personalized learning support and writing assistance, alongside genuine unease about accuracy, ethics, and long-term effects on their own development as learners.

“In one of the largest studies of student attitudes toward generative AI to date, most students reported broadly positive views of these tools for personalized learning support and writing assistance, alongside genuine unease about accuracy, ethics, and long-term effects on their own development as learners (Chan & Hu, 2023).”

• cites Chan & Hu, 2023 sourceNot checked yet

  Not checked in the free run.

May combine two claims.

An exploratory factor analysis supported the twelve-construct structure, with every item loading most strongly on its theorized construct and no meaningful cross-loadings.

“An exploratory factor analysis (reported in Section 4.2) supported the twelveconstruct structure, with every item loading most strongly on its theorized construct and no meaningful cross-loadings.”

Probably not a substantial claim: may not pass S2 (checkable).

Building students' confidence in their own independent academic ability could reinforce formal policy efforts.

“The strong link between Behavioral Control and both lower misuse and higher compliance suggests that building students' confidence in their own independent academic ability could reinforce formal policy efforts.”

May combine two claims.

Academic Pressure's negative association with Compliance Intention points beyond the fraud triangle toward Agnew's general strain theory, which holds that strain erodes the social bonds and commitments that sustain conformity.

“Its negative association with Compliance Intention, however, points beyond the fraud triangle toward Agnew's (1992) general strain theory, which holds that strain not only creates a motive for rule-breaking but also erodes the social bonds and commitments that would otherwise sustain conformity.”

• cites Agnew (1992) sourceNot checked — no full text of the source

• AssumptionRests on an assumption: Self-reported AI Misuse accurately reflects students' actual unauthorized AI use.

May combine two claims.

A maximum-likelihood extraction fixed at twelve factors with varimax rotation produced a clean simple structure: every item's highest loading fell on its theorized construct, primary loadings ranged from .56 to .72, and no item cross-loaded above .32 on any other factor.

“A maximum-likelihood extraction fixed at twelve factors with varimax rotation produced a clean simple structure: every item's highest loading fell on its theorized construct (Table 3), primary loadings ranged from .56 to .72, and no item cross-loaded above .32on any other factor.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Participation was entirely voluntary, and respondents were shown a description of the study's purpose, its anonymity, and their ability to stop at any point without consequence before the questionnaire began.

“Participation was entirely voluntary, and respondents were shown a description of the study's purpose, its anonymity, and their ability to stop at any point without consequence before the questionnaire began.”

Probably not a substantial claim: may not pass S2 (checkable).

Misuse Acceptability and Moral Rationalization point toward institutions engaging students' ethical reasoning about acceptable AI use directly rather than relying on technical and administrative controls alone.

“The roles played by Misuse Acceptability and Moral Rationalization point the same direction, toward institutions engaging students' ethical reasoning about acceptable AI use directly rather than relying on technical and administrative controls alone.”