Page 535
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue IV, April 2026
From Macro Warnings to Micro Risks: Identifying Generative AI
Risks in University Ideological and Political Education
Yanhua Zhong
*
, Baoquan Xie, Yufeng Zou
Institute of Marxism, Ganzhou Polytechnic, Ganzhou, Jiangxi, 341000, China
*Corresponding Author
DOI:
https://doi.org/10.51583/IJLTEMAS.2026.150400048
Received: 12 April 2026; Accepted: 17 April 2026; Published: 06 May 2026
ABSTRACT
The rapid integration of Generative Artificial Intelligence (GAI) into university ideological and political
education has created new pedagogical opportunities while simultaneously introducing significant risks. While
existing research has produced valuable macro level warnings concerning ideological security, discursive ethics,
and teacher authority erosion, these warnings remain largely at the level of principle based alerts. They lack the
granularity needed for frontline teachers to recognize and respond to specific risks in actual teaching scenarios.
This study addresses this gap by shifting the analytical focus from macro warnings to micro risks. Employing a
mixed methods approach that includes a survey of 500 ideological course instructors, semi structured interviews
with teachers and students, and classroom observations across multiple universities in Jiangxi Province, the study
systematically identifies risk manifestations across four teaching scenarios: lesson preparation, classroom
instruction, interactive sessions, and assessment activities. The findings reveal concrete, observable, and
intervenable micro risks, such as GAI generated case studies that subtly weaken the Party's leadership narrative,
students uncritically copying AI generated answers, and automated grading systems that fail to detect
ideologically problematic statements. A preliminary micro risk taxonomy is developed to make these risks visible,
nameable, and actionable. The study further identifies three causal logics underlying these risks: technological
bias embedded in training data, institutional absence of review mechanisms, and cognitive blind spots among
teachers who mistakenly trust GAI neutrality. Theoretically, this study shifts the field from macro warnings to
micro identification. Practically, the micro risk taxonomy provides a diagnostic tool that frontline teachers and
administrators can use to recognize and respond to GAI risks in their daily work.
Keywords: Generative AI, risk identification, micro risks, ideological and political education, higher education
INTRODUCTION
The rapid integration of Generative Artificial Intelligence (GAI), exemplified by large language models such as
ChatGPT and DeepSeek, into university education has opened new pedagogical possibilities. In the specific
context of ideological and political education in Chinese universities, GAI offers promising opportunities for
personalized learning, intelligent tutoring, and dynamic content generation. However, alongside these
opportunities, the deployment of GAI in this value-laden educational domain has raised significant concerns
regarding content security, discursive ethics, teacher authority, and data privacy. Recent global reviews have
documented the rapid proliferation of GAI policies in higher education institutions, highlighting both
Page 536
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue IV, April 2026
opportunities and governance challenges (Jin et al., 2025).
Existing research has produced valuable macro-level warnings about the risks of GAI in ideological education.
Scholars have identified three dominant categories of risk: ideological security risks, such as the potential
weakening of the Party’s leadership narrative through algorithmically generated content (Yu, 2025); ethical and
discursive risks, including the spread of historical nihilist expressions and the erosion of teacher authority as a
discursive gatekeeper (Hu, 2025); and subjectivity erosion risks, where both teachers and students may fall into
uncritical reliance on AI outputs (Yu, 2025; Meng & Yao, 2025). While these warnings are theoretically
important and practically relevant, they share a common limitation: they remain largely at the level of principle-
based alerts, lacking the granularity needed for frontline teachers to recognize and respond to specific risks in
actual teaching scenarios.
This gap between macro warnings and micro practices is not merely academic. For a teacher preparing a lesson
on "Reform and Opening-up," a macro warning about "ideological security risks" offers little guidance on how
to detect whether a GAI-generated case study subtly weakens the leadership narrative. For a student engaging
with an AI-powered Q&A tool, a macro warning about "subjectivity erosion" does not help the instructor identify
when the student is uncritically copying AI-generated answers. The absence of a micro-level, scenario-specific
risk identification framework leaves teachers without actionable diagnostic tools and leaves institutions without
clear governance protocols.
This study addresses this critical gap by shifting the analytical focus from abstract principles to concrete,
observable, and intervenable micro-risks embedded in daily pedagogical practices. Specifically, this research
poses the following questions: (1) What specific risk manifestations emerge when teachers and students use GAI
across different teaching scenarios in ideological and political courses? (2) How can these micro-risks be
systematically identified, named, and categorized? (3) What causal logics underlie the generation of these risks?
Employing a mixed-methods approach that includes a survey of ideological course instructors, semi-structured
interviews with both teachers and students, and classroom observations across multiple universities in Jiangxi
Province, this study systematically identifies risk manifestations across four key teaching scenarios: lesson
preparation, classroom instruction, interactive sessions, and assessment activities. Based on empirical data, the
study proposes a preliminary micro-risk taxonomy for GAI in ideological education, moving the field from
macro warnings toward an empirically grounded, operationally useful framework for risk identification.
The remainder of this paper is structured as follows. Section 2 reviews the existing literature on GAI risks in
ideological education, highlighting the macro-to-micro gap. Section 3 presents the analytical framework and
research methodology. Section 4 reports the empirical findings, including specific risk manifestations across the
four teaching scenarios and a preliminary risk taxonomy. Section 5 discusses the theoretical and practical
implications of shifting from macro warnings to micro identification. Section 6 concludes with limitations and
directions for future research.
LITERATURE REVIEW
Three dominant macro level risk discourses
The existing literature on GAI in ideological and political education has produced three dominant categories of
macro level risk warnings. A systematic review of GAI ethics in higher education has identified recurring themes
of algorithmic bias, data privacy, and pedagogical role transformation (Barus et al., 2025). These categories,
while theoretically important, collectively illustrate the macro level orientation that this study seeks to move
Page 537
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue IV, April 2026
beyond.
The first category concerns ideological security risks. Scholars have warned that GAI, particularly large language
models trained on diverse and often unregulated datasets, may generate content that weakens or distorts official
narratives. Yu (2025) argues that algorithmic recommendations can reinforce historical nihilist tendencies by
presenting decontextualized or one sided accounts of sensitive historical events. Similarly, Hu (2025) points out
that GAI generated content may inadvertently dilute the leadership narrative of the Party when producing case
studies on topics such as “Reform and Opening up” or “Common Prosperity.” These risks are not merely
hypothetical. The underlying concern is that GAI systems, optimized for coherence and user engagement rather
than political accuracy, may produce outputs that are factually correct but ideologically misaligned.
The second category focuses on ethical and discursive risks. Hu (2025) provides a systematic analysis of how
GAI reshapes the discursive ecology of ideological courses. He identifies three core dilemmas: the dissolution
of teacher authority as algorithmic recommendations dilute the teacher's role as a discursive gatekeeper, the
polarization of student audiences as personalized recommendation systems reinforce cognitive echo chambers,
and discursive disorder as GAI generated content struggles to balance political correctness with factual accuracy.
This analysis draws on Habermas's theory of communicative action, which emphasizes truth, sincerity, and
legitimacy as preconditions for productive discourse. When GAI generated content lacks these qualities, the
entire discursive foundation of ideological education is threatened.
The third category addresses subjectivity erosion risks. Yu (2025) and Meng and Yao (2025) have documented
how both teachers and students may develop uncritical dependence on GAI outputs. For teachers, the risk is
becoming a “technical operator” who merely delivers AI generated content without critical mediation. For
students, the risk is falling into “cognitive outsourcing,” where they accept AI generated answers without
questioning their validity or underlying assumptions. Selwyn (2019) has similarly warned in a broader
educational context that AI tools, if not carefully governed, can lead to the commodification of education and
the dehumanization of teacher student relationships. Williamson (2017) adds that learning analytics technologies
risk reducing students to “data points,” undermining their integrity as moral agents.
While these three categories provide a useful starting point, they remain largely descriptive rather than
analytically integrated. Few studies have examined how these risk categories interact or reinforce each other.
For instance, ideological security risks may exacerbate subjectivity erosion when students uncritically accept
biased AI outputs, yet this intersection remains underexplored in the existing literature.
Why macro warnings are insufficient
Despite their theoretical value, macro level warnings share a critical limitation. They remain at the level of
principle based alerts and lack the granularity needed for frontline teachers to recognize and respond to specific
risks in actual teaching scenarios.
This limitation manifests in three specific ways. First, risk identification is abstract rather than concrete. Existing
studies warn about “algorithmic bias” or “ideological security risks” but rarely specify what these risks look like
when a teacher is preparing a lesson, facilitating a classroom discussion, or evaluating student assignments.
Second, governance recommendations are aspirational rather than operational. Scholars commonly call for
“strengthening value rationality,” “enhancing ethical standards,” or “improving digital literacy” (Meng & Yao,
2025; Hu, 2025), but few provide concrete tools that teachers can directly use, such as checklists for reviewing
AI generated content or question banks for guiding student critique. Third, the research perspective is technology
centered rather than teaching centered.
Page 538
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue IV, April 2026
Much of the existing literature focuses on the technical logic of GAI systems themselves, treating ideological
education as a generic application context rather than a distinctive value laden domain with its own pedagogical
logic and political imperatives. Large scale studies on the ethical and societal impacts of GAI in higher education
have similarly noted a tendency to prioritize technical affordances over pedagogical contexts (Mak et al., 2025).
The call for micro level risk research
A small but growing body of literature has begun to call for more micro level, context sensitive approaches to
GAI risk research. In the field of AI ethics, Floridi (2019) has argued that digital ethics must move from
principles to practices, embedding ethical requirements into technical design rather than leaving them as abstract
aspirations. Bostrom and Yudkowsky (2018) have warned about the danger of “instrumental convergence,”
where an AI system optimized for a narrow goal may pursue that goal in ways that violate broader human values.
These insights, while developed primarily in the context of general AI safety, have direct relevance for
ideological education. If GAI systems are optimized for student engagement or content coherence without
explicit value alignment, they may generate outputs that are pedagogically effective but ideologically
problematic.
In the specific context of Chinese ideological education, the call for micro level research is particularly urgent.
The macro warnings currently available do not answer practical questions such as: When a teacher uses GAI to
generate a case study on “Reform and Opening up,” what specific wording patterns should trigger a manual
review? When a student uses a GAI powered Q&A tool, how can the instructor distinguish between legitimate
help and uncritical copying? When an institution considers adopting a GAI system for automated grading, what
privacy safeguards must be in place? Table 1 provides a structured comparison between existing macro level risk
research and the micro level approach taken by this study.
Table 1. Summary of Macro Level Risk Research and the Micro Level Gap
Dimension
What Existing Research Has Done
(Macro Level)
What This Study Does
(Micro Level)
Risk
identification
Provides principle based warnings
(e.g., “ideological security risks,”
“subjectivity erosion”)
Identifies concrete, observable, and
intervenable risk manifestations across
specific teaching scenarios
Risk categories
Offers three broad categories:
ideological security, ethical and
discursive, and subjectivity erosion
Develops a scenario specific micro risk
taxonomy (lesson preparation, instruction,
interaction, assessment)
Governance
focus
Calls for strengthening value
rationality, enhancing ethical standards,
and improving digital literacy
Provides operational diagnostic tools
including a risk checklist and identification
framework
Research
perspective
Technology centered, treating
ideological education as a generic
application context
Teaching centered, grounded in the
pedagogical logic and political imperatives of
ideological courses
Empirical basis
Largely theoretical or conceptual with
limited classroom level data
Empirically grounded in surveys, interviews,
and classroom observations across multiple
universities
Page 539
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue IV, April 2026
Research gap and contribution of this study
The literature review reveals a clear research gap. Existing studies have produced valuable macro level warnings
about GAI risks in ideological education, but they have not systematically identified the specific, observable,
and intervenable micro risks that emerge in actual teaching scenarios. Consequently, teachers lack actionable
diagnostic tools, and institutions lack clear governance protocols.
This study addresses this gap by shifting the analytical focus from macro warnings to micro identification. The
contribution is twofold. Theoretically, the study develops a preliminary micro risk taxonomy for GAI in
ideological education, moving the field from abstract principle based alerts to empirically grounded, scenario
specific risk categories. Practically, the study provides a diagnostic tool that frontline teachers and administrators
can use to recognize, name, and respond to GAI risks in their daily work. This shift from macro to micro is a
necessary step toward developing governance strategies that serve rather than alarm frontline educators.
RESEARCH METHODOLOGY
Research design
This study employs a mixed methods research design that combines quantitative and qualitative approaches. The
rationale for using a mixed methods design is twofold. First, the research questions require both breadth (to
understand the prevalence of GAI risks across a larger population) and depth (to capture the specific
manifestations of risks in authentic classroom contexts). Second, the exploratory nature of this study, which
seeks to identify previously unnamed micro risks, benefits from qualitative data that can capture unexpected
phenomena while quantitative data provides a sense of scale. Table 2 below summarizes the key components of
the research design and data collection.
Table 2. Summary of Research Design and Data Collection
Research Component
Description
Research design
Mixed methods (quantitative + qualitative)
Quantitative method
Survey questionnaire
Quantitative sample
500 ideological course instructors in Jiangxi Province
Qualitative method 1
Semi structured interviews
Qualitative sample 1
20 ideological course instructors
Qualitative sample 2
25 students who have used GAI assisted learning
Qualitative method 2
Non-participatory classroom observations
Number of observations
2 lessons per participating teacher
Additional materials
GAI generated screenshots, teaching reflections, and negative cases
Data analysis tools
NVivo for qualitative (thematic analysis); SPSS for quantitative (descriptive
statistics)
Quantitative component
The quantitative component consists of a survey questionnaire distributed to ideological course instructors in
universities across Jiangxi Province. The target sample size is 500 respondents. The survey focuses on three core
questions: whether and how frequently instructors use GAI for teaching, what specific problems they have
encountered, and what types of support they most need.
Page 540
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue IV, April 2026
The survey instrument was developed based on the literature review and pilot tested with five instructors for
clarity and relevance. The final version includes closed ended questions with Likert scale responses and multiple
choice items. Descriptive statistics, including frequency distributions and percentages, are used to analyze the
survey data.
Qualitative component
The qualitative component includes three data collection methods: semi structured interviews, non participatory
classroom observations, and collection of supplementary materials such as GAI generated screenshots and
teaching reflections.
Semi structured interviews
Interviews are conducted with two groups of participants. The first group includes 20 ideological course
instructors with varying years of teaching experience, professional titles, and frequencies of GAI use. The
interview protocol explores their practical experiences, concerns, coping strategies, and institutional
expectations regarding GAI use. The second group includes 25 students who have experienced GAI assisted
instruction. The student interview focuses on their trust in AI generated content, critical awareness, and privacy
perceptions. All interviews are audio recorded with participant consent and transcribed verbatim.
Classroom observations
Non participatory classroom observations are conducted in 2 lessons per participating teacher. The purpose is to
document how GAI is actually used in real time, how teachers mediate AI generated content, and how students
respond. Potential risk moments, such as when GAI generated content contains value biases that go unnoticed,
are recorded as observation notes.
Supplementary materials
Participating teachers are invited to share GAI related teaching materials, including screenshots of AI generated
content, teaching reflections on GAI use, and examples of problematic outputs. These materials serve as
additional data sources for identifying risk manifestations.
Data analysis
Qualitative data from interviews, observations, and supplementary materials are analyzed using thematic
analysis with the assistance of NVivo software. The analysis follows a three stage coding process: open coding
to identify initial themes, axial coding to group themes into categories, and selective coding to integrate
categories into a coherent framework.
Quantitative data from the survey are analyzed using SPSS. Descriptive statistics, including frequency
distributions and means, are used to summarize the prevalence of different risk types and support needs. The
quantitative results are used primarily to validate and contextualize the qualitative findings rather than as the
main analytical driver.
Ethical considerations
This study has received ethical approval from the affiliated university. All participants are informed of the
research purpose and their rights, including the right to withdraw at any time. Informed consent is obtained from
all participants. For classroom observations, both the instructor and the students are informed in advance. All
Page 541
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue IV, April 2026
data are anonymized, with personal identifiers removed before analysis.
RESULTS
This section presents the empirical findings organized around the four teaching scenarios examined in this study:
lesson preparation, classroom instruction, interactive sessions, and assessment activities. The findings are drawn
from survey data, interview transcripts, classroom observation notes, and supplementary materials collected
from participating teachers and students.
Risk manifestations across four teaching scenarios
Lesson preparation risks
In the lesson preparation scenario, teachers reported using GAI to generate case studies, historical summaries,
and discussion questions. While GAI significantly reduced preparation time, it also introduced subtle but
consequential risks. Several teachers noted that GAI generated case studies on politically sensitive topics such
as “Reform and Opening up” tended to emphasize economic achievements while downplaying the leadership
narrative of the Party. One teacher explained, “The AI generated text was factually correct, but something felt
off. It told the story of economic growth without mentioning who led the reform.” Similarly, GAI produced
historical summaries occasionally contained decontextualized accounts of sensitive events, presenting them as
neutral facts rather than as part of a coherent ideological framework.
Classroom instruction risks
During classroom instruction, teachers displayed GAI generated content on screens or used GAI to provide real
time examples. The most frequently reported risk was the presence of historically nihilist expressions. For
instance, when a teacher used GAI to generate a description of a revolutionary hero, the output described the
hero’s personal struggles in detail but omitted the broader political significance of their actions. Another teacher
reported that GAI generated definitions of “Common Prosperity” overemphasized efficiency and market
mechanisms while underemphasizing equity and redistribution. A third risk involved implicit bias, where GAI
produced examples that presented Western political systems as neutral benchmarks for comparison without
acknowledging their ideological assumptions.
Interactive ssession risks
In interactive sessions, students used GAI powered Q&A tools to assist with in class discussions and assignment
preparation. The most concerning risk was students copying AI generated answers without critical scrutiny.
Interview data revealed that many students trusted GAI outputs as authoritative, rarely questioning their validity
or underlying assumptions. One student admitted, “If ChatGPT gives me an answer, I assume it is correct. I do
not have time to fact check everything.” This phenomenon, which can be described as cognitive outsourcing,
represents a form of subjectivity erosion where students delegate critical thinking to the AI. Recent research on
GAI in ideological education has similarly identified this erosion of educational object subjectivity as a core
ethical concern (Yan, 2025). Additionally, classroom observations recorded instances where GAI chatbots
provided vague or evasive responses to politically sensitive student questions, leaving students confused rather
than informed.
Assessment activity risks
In the assessment scenario, teachers reported using GAI to assist with grading and feedback, while students used
Page 542
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue IV, April 2026
GAI to generate essay submissions. Teachers noted that automated grading systems sometimes penalized
politically appropriate but linguistically non standard expressions, suggesting algorithmic bias in how content is
evaluated. Conversely, automated feedback systems occasionally failed to detect ideologically problematic
statements in student submissions, creating a content blind spot. On the student side, several teachers expressed
concern about academic integrity, noting that some students submitted GAI generated essays without proper
disclosure or critical adaptation. As one teacher put it, “The students are not cheating in the traditional sense, but
they are also not learning. The AI does the thinking for them.”
Summary of micro risk taxonomy
To provide a structured overview of the empirical findings, Table 3 presents a micro risk taxonomy that
organizes specific risk manifestations by teaching scenario and risk type. This taxonomy moves beyond macro
level warnings by naming concrete, observable, and intervenable risks that frontline teachers can recognize in
their daily practice.
Table 3. Micro Risk Taxonomy of GAI in Ideological and Political Education
Teaching Scenario
Risk Type
Lesson preparation
Content security
Ideological
misalignment
Discursive risk
Classroom
instruction
Ideological
security
Value distortion
Implicit bias
Interactive sessions
Subjectivity
erosion
Cognitive
outsourcing
Discursive
disorder
Assessment
activities
Algorithmic bias
Content blind spot
Academic
integrity
Causal logics underlying micro risks
Beyond identifying specific risk manifestations, this study also examined why these risks occur. The data suggest
three interconnected causal logics.
Page 543
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue IV, April 2026
Technological bias
The first causal logic concerns the training data and optimization objectives of GAI systems. Many GAI models
are trained on diverse but predominantly Western datasets, which embed certain political and cultural
assumptions. When these models are applied to the Chinese ideological education context, they may produce
outputs that are statistically coherent but ideologically misaligned. Additionally, GAI systems are optimized for
user engagement and content fluency rather than political accuracy, creating a built in tension with the
requirements of ideological education.
Institutional absence
The second causal logic involves the lack of institutional policies and review mechanisms. Most participating
teachers reported that their universities had no clear guidelines on acceptable GAI use in teaching. Without
institutional standards, individual teachers are left to navigate GAI risks on their own, leading to inconsistent
practices. One teacher explained, “I do not know what is allowed and what is not. I just try to be careful, but I
have no official guidance.”
Cognitive blind spots
The third causal logic relates to cognitive blind spots among teachers. Several interview participants expressed
misplaced trust in GAI neutrality, assuming that AI generated content is inherently objective or value free. This
assumption overlooks the embedded values and biases in GAI systems. As one teacher admitted, “I never thought
that AI could have a political bias. I assumed it was just pulling facts from the internet” (Interview, Teacher #12,
2025). This cognitive blind spot leaves teachers vulnerable to unrecognized ideological misalignments in GAI
outputs.
DISCUSSION
This section discusses the theoretical and practical implications of shifting the analytical focus from macro
warnings to micro risks in the study of GAI in ideological and political education. It also addresses the study's
limitations and proposes directions for future research.
The macro to micro shift as a necessary paradigm turn
The findings of this study demonstrate that macro level warnings, while theoretically important, are insufficient
for frontline teachers who need to recognize and respond to specific GAI risks in real time. By identifying
concrete risk manifestations across four teaching scenarios, this study provides empirical evidence that micro
risks are not merely smaller versions of macro risks. Rather, they represent a distinct analytical level that requires
its own vocabulary, taxonomy, and diagnostic tools.
This macro to micro shift parallels similar developments in other fields. In public health, for example, macro
level warnings about "health risks" are useful but must be supplemented by micro level screening tools that help
clinicians identify specific symptoms. In cybersecurity, general warnings about "data breaches" are less useful
than specific checklists that help users recognize phishing attempts. Similarly, in GAI risk research, moving from
principle based alerts to scenario specific risk identification is a necessary step toward developing operable
governance strategies.
The theoretical contribution of this shift is twofold. First, it challenges the assumption that macro level warnings
naturally translate into micro level practices. The gap documented in this study suggests that translation is neither
Page 544
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue IV, April 2026
automatic nor straightforward. Second, it provides an empirically grounded micro risk taxonomy that can serve
as a foundation for future theory building. Rather than starting from abstract principles, future research can build
on the specific risk types identified in this study to develop more nuanced models of GAI risk generation and
propagation.
Redefining teacher professionalism in the GAI era
The findings also have implications for how teacher professionalism is understood in the context of GAI. Much
of the existing literature has focused on the threat that GAI poses to teacher authority, framing the relationship
between teachers and AI as inherently competitive. This study suggests a different framing. The teachers in this
study were not passive victims of technological disruption. They were actively developing strategies to review,
critique, and guide GAI use, even in the absence of institutional support.
This finding aligns with the concept of teacher agency, which emphasizes the active role that teachers play in
shaping technology use rather than being shaped by it. In the GAI era, teacher professionalism may need to be
redefined to include not only content knowledge and pedagogical skill but also what might be called algorithmic
discernment: the ability to identify value misalignments in AI generated content, to scaffold student critique of
AI outputs, and to make contextual judgments about when and how to use GAI in politically sensitive teaching
environments.
This redefinition has practical implications for teacher training and professional development. Pre service and
in service training programs should move beyond generic digital literacy modules to include scenario based
training on GAI risk identification. Teachers need opportunities to practice reviewing GAI generated content, to
discuss edge cases with peers, and to develop shared norms for acceptable GAI use.
From individual coping to institutional governance
A third implication concerns the level at which GAI risks are best governed. The findings reveal that most
teachers are currently coping with GAI risks on an individual basis, with little institutional guidance or support.
This individual coping strategy is unsustainable and inequitable. Teachers with higher digital literacy or more
institutional support are better equipped to manage risks, while others are left vulnerable.
The data suggest a need for institutional governance mechanisms that operate at three levels. At the teacher level,
institutions should provide practical tools such as risk checklists, review protocols, and teaching scripts that help
teachers identify and respond to common GAI risks. At the course level, institutions should support curriculum
integration efforts that embed AI literacy and critical AI use into existing course structures rather than treating
them as add ons. At the institutional level, universities should develop clear policies on acceptable GAI use,
establish review mechanisms for high risk applications such as automated grading, and create reporting channels
for problematic GAI outputs. This three level governance framework aligns with recent institutional intervention
models proposed in the literature (Jayasinghe et al., 2026).
Notably, these governance mechanisms need not be burdensome or expensive. Many of the tools that teachers
requested in this study, such as a one page risk checklist or a set of example teaching scripts, are low cost
interventions that could be developed and disseminated with modest resources. The barrier is not primarily
financial but organizational, requiring institutions to prioritize GAI governance as part of their teaching quality
assurance systems.
While this study focuses on risk identification, it is important to note that identification is only the first step
toward mitigation. Future research should build on the taxonomy developed here to design and test specific
Page 545
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue IV, April 2026
intervention strategies. Potential mitigation approaches include teacher training modules on AI output review,
student workshops on critical AI literacy, and institutional policy frameworks for GAI governance. The
taxonomy provides a diagnostic foundation for these interventions by specifying what risks need to be addressed.
Limitations
This study has several limitations that should be acknowledged. First, the empirical data were collected
exclusively from universities in Jiangxi Province. While this regional focus allowed for in depth data collection,
it limits the generalizability of the findings to other provinces or countries with different political, cultural, or
technological contexts. Second, the study relied primarily on self reported data from interviews and surveys,
which may be subject to social desirability bias or recall limitations. Classroom observations helped mitigate
this limitation, but observational data were limited to one or two lessons per teacher. Third, the rapid pace of
GAI development means that specific risk manifestations may change as new models and applications emerge.
The taxonomy proposed in this study should be treated as a preliminary framework that requires regular updating
rather than a fixed classification. Fourth, this study focused on risk identification rather than risk mitigation or
intervention testing. While the taxonomy provides a diagnostic foundation, future research is needed to develop
and test specific interventions.
Future research directions
The findings of this study open several avenues for future research. First, the micro risk taxonomy should be
validated and refined through larger scale studies across multiple provinces and institutional types. Quantitative
surveys could assess the prevalence of different risk types and identify teacher or institutional characteristics
associated with higher risk exposure. Second, intervention research is needed to test the effectiveness of specific
governance tools, such as risk checklists or training programs. Randomized controlled trials or quasi
experimental designs could compare outcomes between teachers who receive specific interventions and those
who do not. Third, comparative research across different countries or political systems would help clarify which
aspects of GAI risk are universal and which are context specific. Such research would contribute to a more
nuanced understanding of the relationship between GAI, ideology, and education. Fourth, longitudinal research
could track how GAI risks evolve over time as technologies change and as teachers and students develop new
coping strategies. Such research would help anticipate emerging risks before they become widespread.
CONCLUSION
This study addressed the gap between macro level warnings and micro level practices in research on Generative
Artificial Intelligence (GAI) in university ideological and political education. Drawing on survey data,
interviews, and classroom observations across multiple universities in Jiangxi Province, the study identified
specific, observable, and intervenable risk manifestations across four teaching scenarios: lesson preparation,
classroom instruction, interactive sessions, and assessment activities. A preliminary micro risk taxonomy was
developed to make these risks visible, nameable, and actionable. Theoretically, the study shifts the analytical
focus from macro warnings to micro identification. Practically, the taxonomy provides a diagnostic tool for
frontline teachers and administrators.
The study has several limitations, including its regional focus and its reliance on self-reported data, and its
emphasis on risk identification rather than intervention testing. Additionally, given the exploratory phase of this
research and the novelty of the micro-risk taxonomy, the quantitative analysis was primarily descriptive. This
approach is appropriate for establishing foundational patterns and frequencies. Future research, building on this
taxonomy, should employ inferential statistical techniques (e.g., regression analysis) to examine the predictive
Page 546
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue IV, April 2026
relationships between teacher characteristics, institutional contexts, and the prevalence of specific micro-risks.
The integration of GAI into ideological education is unlikely to reverse. The question is not whether to use these
technologies, but how to govern their risks intentionally. This study has taken a first step toward answering that
question.
REFERENCES
1. Barus, O., Hidayanto, A. N., & Eitiveni, I. (2025). Mapping generative AI's ethical issues in higher
education: A FELT-guided systematic review. Polyglot: Jurnal Ilmiah,
21(2). https://doi.org/10.19166/pji.v21i2.10020
2. Bostrom, N., & Yudkowsky, E. (2018). The ethics of artificial intelligence. In Artificial intelligence
safety and security (pp. 57-69). Chapman and Hall/CRC.
3. Dai, J. P., & Qin, Y. Y. (2023). The ideological risks of generative artificial intelligence such as ChatGPT
and its response. Journal of Chongqing University (Social Science Edition), 29(5), 101-110.
4. Floridi, L. (2019). Translating principles into practices of digital ethics: Five risks of being
ethical. Philosophy & Technology, 32(2), 185-193. https://doi.org/10.1007/s13347-019-00354-x
5. Habermas, J. (1984). The theory of communicative action (T. McCarthy, Trans.). Beacon Press. (Original
work published 1981)
6. Hu, G. (2025). Discursive ethical risks and governance paths of ideological and political courses in
universities in the era of generative artificial intelligence. Heilongjiang Researches on Higher Education,
43(9). (forthcoming)
7. Jin, Y., Yan, L., Echeverria, V., Gašević, D., & Martinez-Maldonado, R. (2025). Generative AI in higher
education: A global perspective of institutional adoption policies and guidelines. Computers and
Education: Artificial Intelligence, 8, 100348. https://doi.org/10.1016/j.caeai.2024.100348
8. Mak, J., Nakatumba-Nabende, J., Clear, T., Clear, A., Albluwi, I., Andrei, O., Angeli, L., MacNeil, S.,
Oyelere, S. S., Rattigan, M. H., Sheard, J., & Zhu, T. (2025). Navigating the ethical and societal impacts
of generative AI in higher computing education (arXiv:2511.15768v1).
arXiv. https://doi.org/10.48550/arXiv.2511.15768
9. Meng, Q. P., & Yao, H. X. (2025). The internal mechanism, risks and countermeasures of AI-driven
teaching reform in university ideological and political courses. Modern Distance Education Research,
37(3). (forthcoming)
10. Selwyn, N. (2019). Should robots replace teachers? AI and the future of education. Polity Press.
11. Jayasinghe, S., Gamage, K. A., Yang, D., Cheng, C., Disanayake, C., & Apeji, U. D. (2026). Six
Institutional Intervention Areas to Support Ethical and Effective Student Use of Generative AI in Higher
Education: A Narrative Review. Education Sciences, 16(1),
137.. https://doi.org/10.3390/educsci16010137
12. Wang, S. J., & Zhang, Y. (2024). The basic logic and contradiction adaptation of generative AI
intervening in ideological and political education: From ChatGPT to GPT-4o. Ideological Education
Research, 2024(12), 52-58.
13. Williamson, B. (2017). Big data in education: The digital future of learning, policy and practice. SAGE
Publications.
14. Yan, R. F. (2025). Ethical risks and resolution paths of generative AI empowering ideological and
political education: An investigation based on the perspective of educational object subjectivity. Marxist
Studies Network. http://marxism.cass.cn/zzjy/202510/t20251029_5921859.shtml
15. Yu, Y. (2025). Risk prevention and practical exploration of AI-empowered ideological and political
course teaching in universities. Journal of Langfang Normal University (Social Sciences Edition), 41(2).
16. Yue, Q., & Chen, M. Z. (2025). Coupled mechanisms, risk challenges, and ecological reconstruction of
Page 547
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue IV, April 2026
GenAI enabling ideological and political education. Journal of Zhengzhou University of Light Industry
(Social Science Edition), 26(5), 59-67. https://doi.org/10.12186/2025.05.008
17. Ludi, Z. H. A. O., & Shengli, Q. I. (2025). Ethical Risks and Prevention of Generative Artificial
Intelligence Empowering Ideological and Political Education for College Students. The journal of
xinyang normal university (philosophy and social science edition), 45(6), 16-24.
18. Zhong, H. (2025). The Opportunities, Challenges and Countermeasures Brought by the Development of
Chat Generative. In Proceedings of the 2025 9th International Seminar on Education, Management and
Social Sciences (ISEMSS 2025) (p. 281). Springer Nature. https://doi.org/10.2991/978-2-38476-462-
4_32