The OBM industry is experiencing a major transformation due to AI adoption. Unlike OEM or ODM models, OBM firms not only produce goods but also build their own brands, which requires constant innovation and customer engagement [30][51]-[53]. Many manufacturing firms use AI to enhance product development, streamline production, and optimize supply chains [12][27][54][55]. Therefore. Employees remain central to product and process innovation, which underpin brand value. When AI adoption undermines job security or reduces employees’ sense of ownership, their motivation to take creative risks and innovate declines [56]. Studies show that perceiving AI as a threat lowers engagement and stifles innovation. To counter this, recent research emphasizes collaborative intelligence, using AI to support, rather than replace, human creativity [57]-[60]. Understanding how AI acts as both an enabler and a barrier to innovation is vital. Analyzing this dual impact through established theories can provide deeper insight into AI’s role in shaping innovative behavior at work.

Beyond automating routine tasks, AI in OBM firms enables intelligent decision-making, predictive analytics, complex problem-solving, and generative design [12][24][61][62]. Through technologies such as machine learning, computer vision, and natural language processing, AI systems give employees practical support to innovate: they accelerate product design and prototyping, enable real-time market insights, and optimize production workflows, freeing employees to focus on higher-value creative activities. By reducing cognitive workload, AI enhances both efficiency and employee well-being. When AI supports employees with repetitive tasks, they report higher job satisfaction and engagement [63]. If employees see AI as a helpful tool, they become more engaged and adaptable [64].

The TAM explains why AI adoption can drive employee innovation. According to TAM, employees are more likely to use new technologies they see as practical and easy to use [65]. In OBM firms, if employees believe AI enhances their effectiveness or creativity, they will integrate it into their workflow, increasing both exploratory (trying new ideas) and exploitative (improving existing ideas) innovation [66]. For example, employees can use generative AI for prototyping and AI analytics to spot market opportunities, boosting innovation overall [67]. Empirical research shows that combining generative AI with human creativity improves innovative performance [41][68]-[70]. When employees see AI as a partner, they receive better cognitive support and make faster decisions, which enhances innovative behavior [40][71]. However, AI adoption alone does not guarantee more innovation. If employees view AI as a competitor instead of a collaborator, for instance, when algorithmic decisions override human input, they are less likely to take initiative [33]. This duality underscores the importance of examining AI’s concurrent enabling and suppressive effects on employee behavior.

AI technologies enhance idea development, but over-reliance on them reduces employees’ sense of ownership and autonomy, which in turn limits innovation [46]. Heavy dependence on AI also raises job security concerns [72], and shifts how human contributions are valued, directly undermining innovative activity. In OBM firms, employees play a central role in product and brand development. The excessive automation of creative and decision-making processes can make workers feel replaceable and undervalued, a pattern documented across organizational contexts where AI encroaches on human judgment and expertise [73]. When AI determines most product characteristics or heavily influences strategic choices, employees perceive that their expertise is disregarded. Empirical studies confirm these concerns. Employees exposed to advanced AI and robotics report lower job security, reduced organizational commitment, and heightened anxiety about the future. Brougham and Haar [74] found that greater awareness of STARA was negatively related to organizational commitment and career satisfaction, while positively linked to turnover intentions, cynicism, and depression. Similarly, Bhargava, Bester [33] showed that employees who perceived RAIA implementation in their workplace reported lower job security and reduced confidence in their long-term employability, factors that erode the motivation to take initiative and contribute innovatively.

The COR theory [38] explains why AI-related threats can suppress innovation. Employees seek to protect vital resources such as job security, autonomy, and self-efficacy. When these resources appear at risk, they adopt a defensive posture, prioritizing preservation over exploration. As a result, their willingness to engage in innovative activities declines [45]. Integrating TAM and COR clarifies AI’s double-edged effect: TAM highlights the enabling side when AI is embraced as useful, while COR explains the risks when AI is seen as a threat. Recent research supports this duality: Du and Liu [46] found that employees credited AI with improving focus but also felt it reduced their autonomy, showing that AI adoption brings both opportunities and risks. Several psychological mechanisms can explain why AI adoption may dampen innovative behavior in OBM firms.

Given these contrasting dynamics, a critical question arises: Does AI adoption enable or inhibit innovation in OBM firms? If AI supports employees (as TAM suggests), innovation should increase. But if AI causes resource-loss fears (as COR predicts), innovation will decline. Both effects may happen at once. In OBM, where efficiency and creativity must be balanced, understanding this trade-off is essential. Therefore, we argue that AI adoption significantly affects how employees engage in innovation and seek to test this empirically. Accordingly, we propose:

H1:AI adoptionhas a significant negative impact on employee innovative behavior in OBM firms.

Job security generally refers to an employee’s confidence in the continued stability of their job and the absence of threats to their position [34]. More specifically, it captures employees’ perception of stability and continuity within their roles [75]. Across industries, new technologies like robotics and AI have raised employee concerns about job stability [76]. AI-driven automation has intensified these concerns, as it can replace tasks and create uncertainty about future employment [26]. If organizations do not address these anxieties, employee stress can negatively impact performance [38][77][78]. AI adoption may make employees feel insecure and think AI may replace their jobs [74]. Recent surveys and studies confirm this: a 2024 Pew survey found 52% of U.S. workers are concerned about AI’s impact on their jobs, and one-third believe AI will lead to fewer job opportunities for them in the long run [79]. In fast-changing industries, such as OBM, AI adoption has introduced concerns regarding job displacement and role obsolescence [41][70]. While AI has the potential to enhance productivity and efficiency, it simultaneously raises apprehensions about workforce reductions and shifts in job responsibilities [68][69]. While most research has examined AI’s impact on organizational performance, less attention has been given to its effects on individual employees’ job security and innovative behavior. Since job security strongly influences motivation, risk-taking, and creativity, it is essential to understand its mediating role in the AI-innovation relationship.

AI impacts job security in two major ways. First, by automating routine tasks like design iteration, supply chain forecasting, and optimizing production, AI can make some jobs redundant or outdated [80], fueling worker anxiety. Second, AI adoption demands new skills: employees need digital literacy and the ability to understand algorithmic processes to stay competitive [81]. Without organizational support for upskilling, employees feel more anxious and may resist AI integration [82]. While AI makes workflows more efficient [70], it also increases fears of job loss [41][68].

COR theory helps to explain how AI adoption can affect employees’ sense of job security, and also explains that employees aim to protect valued resources such as job stability [38]. When AI adoption is perceived as a threat to these resources, employees become stressed and defensive [45]. This reduces engagement and commitment at work.

Research in industry fields shows a consistent negative link between AI adoption and job security: in hospitality, for example, AI-induced automation has raised job security concerns, which has introduced lowered job engagement [83]. Efforts to adopt AI and pursue digital transformation have led to job cuts and uncertainty in various industries [82], with manufacturing workers fearing job loss and fewer career opportunities [35]. Overall, AI adoption often weakens employees’ sense of job security, leading to increased stress and reduced motivation. Based on this evidence, we propose:

H2a:AI adoptionnegatively influences job security.

Job security strongly affects employees’ willingness to participate in innovation [84][85]. When employees feel secure in their jobs, they take more risks, explore new ideas, and actively support innovation within their organization [86]. Many researchers have found a clear positive link between job security and employee innovation [87]-[89]. COR theory Hobfoll [38] helps explain this link: employees are more willing to invest time, energy, and creativity in innovation when they feel their jobs are safe [35][83]. According to the TAM, employees who feel secure see new technologies as opportunities to improve their work, not threats [30][65][68].

Empirical studies confirm that job security encourages creativity, problem-solving, and new idea generation [86][90]. On the other hand, those who faced job insecurity were less engaged in innovation [83]. Later, Probst, Chizh [91] confirmed that job security helps build a sense of stability and trust in the workplace. In contrast, those who feel insecure are less likely to take risks or innovate. Therefore, these findings suggest that feeling secure in their jobs encourages employees to be more innovative. Based on this, we propose:

H2b:Job security positively influences employee innovative behavior.

Based on these arguments, job security mediates the relationship between AI adoption and innovative behavior. When AI adoption threatens job security, employees feel anxious and avoid risks, which reduces innovation [35][68][91]. If organizations use AI to support or enhance job security, such as through upskilling or clarifying AI’s supportive role, employees are more likely to adopt AI and innovate [30][83]. According to TAM, perceived usefulness increases willingness to use new technology, and support from the organization boosts innovative behavior [68][69]. COR theory also supports this: protecting job stability encourages employees to invest resources in innovation. When job security is lacking, employees focus on managing insecurity instead of being creative. Therefore, job security can transmit the effects of AI adoption to employee innovative behavior, buffering against the negative and allowing the positive potential of AI to be realized. Therefore, we propose:

H2:Job security mediates the relationship between AI adoption and employee innovative behavior.

Employee well-being refers to the overall mental and emotional condition of an employee, reflecting resilience and the ability to manage workplace stress [20][92]. It plays a crucial role in shaping how employees perceive and react to AI adoption [93]. AI can serve as both an enabler and a stressor, depending on employees’ psychological resources and adaptability [94][95]. According to the COR theory, individuals strive to preserve valued resources such as job security, psychological well-being, and professional stability [38]. When employees perceive AI as a tool that enhances productivity and autonomy, they view it as a valuable resource. This fosters engagement, creativity, and ultimately greater innovation [50][96]. In contrast, when AI is seen as a threat to job stability or a source of workload uncertainty, employees feel their resources are being eroded. This perception generates stress and anxiety, reducing their motivation to innovate [97][98]. In this way, employee well-being moderates the relationship between AI adoption and innovative behavior. Employees with high well-being possess stronger psychological resources, which help them adapt to technological change and use AI as a tool for innovation [99]. Conversely, employees with lower well-being are more vulnerable to AI-related uncertainty and tend to avoid risks, which limits creativity [74][100][101]. COR theory supports this view, explaining that individuals with fewer resources conserve energy by avoiding additional efforts such as innovation [102].

Well-being also shapes how employees connect job security with AI adoption. Concerns about automation, job loss, or increased digital-skill demands heighten insecurity for employees lacking psychological support [103]. From the COR perspective, low well-being amplifies these threats, leading to anxiety and resistance to AI-driven change [104]. In contrast, employees with higher well-being are more likely to interpret AI as an opportunity for growth, adjust their skills, and embrace AI integration without feeling undermined [93]. Thus, employee well-being serves as a protective buffer, softening the negative effects of AI adoption on job security. From this perspective [65], while COR explains how employees protect personal resources [38]. As a result, employees with higher well-being are less threatened by AI and better positioned to leverage it as a resource for growth. Based on these arguments, we propose:

H3: Employee well-being moderates the relationship between AI adoption and employee innovative behavior, such that the negative relationship is weaker (i.e., less harmful to innovation) when employee well-being is high, and stronger (i.e., more harmful) when well-being is low.

H4: Employee well-being moderates the relationship between AI adoption and job security, such that higher well-being reduces the negative impact of AI adoption on job security.

Despite growing interest in AI and innovation, few studies have empirically examined these mediated (job security) and moderated (employee well-being) mechanisms in manufacturing contexts where both creative agility and operational precision are essential. This study addresses this gap in the OBM industry, where understanding these dynamics is critical to leveraging AI in an innovation-driven environment.

This study used a quantitative research design to examine the relationship between AI adoption (AIA) and employee innovative behavior (EIB), with job security (JS) as a mediator and employee well-being (EWB) as a moderator. To reduce common method bias [105][106], data were collected in a two-wave online survey from employees within OBM industry in China. Participants were drawn from R&D, product management, marketing, and project management teams, key functions in AI-driven innovation [107]-[109]. Prior studies show AI in these roles enhances forecasting, automation, market insights, and collaboration [110]-[112], though concerns over job security remain [113]. Data were collected via Credamo, which offers high-quality recruitment, data cleaning, and secure administration. The study adhered to strict ethical guidelines, including informed consent, anonymity, confidentiality, voluntary participation, and verification that respondents were members of product innovation teams to ensure eligibility and data accuracy. This approach aligns with best practices by ensuring diversity, reducing geographic limitations, and maintaining cost efficiency [114]. In the first wave, we collected responses on AI adoption (IV) and job security (mediator). Four weeks later, the same participants completed employee well-being (moderator) and employee innovative behavior (DV) measures. This temporal separation helped reduce potential common method bias between the predictor and outcome variables [105][115].

We targeted employees in OBM firms who worked in innovation-driven roles. To verify eligibility, respondents were first required to confirm that their employer manufactured products under its own brand and was responsible for marketing or commercializing that brand, rather than operating exclusively as an OEM or ODM firm. Respondents were also asked whether they were currently part of a product innovation team, such as R&D, design, product management, marketing management, or project management. Only respondents who satisfied both criteria were allowed to proceed with the survey. A total of 400 survey invitations were distributed, resulting in 330 valid responses in the first wave (82.5% response rate). After four weeks, we sent the questionnaire to the same 330 participants contacted in the first wave, and we received 290 valid responses, resulting in a retention rate of 87.88%. After checking for logic and quality, 39 invalid questionnaires were removed. These included responses with unusually short completion times, failed screening questions, or repetitive answer patterns. A total of 251 valid questionnaires were retained for analysis. The effective response rate was 86.55%. Demographic details are shown in Table 1. The final dataset met the recommended sample size thresholds for PLS-SEM [116].

All variables were measured using validated multi-item scales from previous studies. Responses were recorded on a five-point Likert scale (1 = strongly disagree, 5 = strongly agree). Compared to a seven-point scale, a five-point format reduces confusion over small differences between options. This improves response quality [117][118].

AI adoption: The AI adoption scale was adapted from [119] measure of computer use. We replaced “computer” with “AI” and revised the items based on expert input and the specific goals of this study. In this study, AI adoption refers to the degree to which employees integrate and depend on AI tools in their day-to-day work tasks, a usage-centered conceptualization reflecting how deeply AI has been embedded into each employee’s working practice, rather than a binary indicator of whether AI exists in the organization. This operationalization captures the employee’s lived experience of AI at the task level, consistent with how job-level AI exposure has been measured in prior individual-level studies [68][97]. The measurement scales used were originally developed in English, we conducted a commonly used translation and back-translation procedure for the items to ensure the equivalency of meaning in the Chinese version. The final version includes eight items. One example is, “I need AI to help me do my job.” The Cronbach’s alpha is 0.927.

Job security: The JS was measured using a 10-item scale developed by [75]. An example item is: “I’ll be able to keep my present job as long as I wish.” The Cronbach’s alpha is 0.848.

Employee well-being was assessed using the well-being scales proposed by [92], which measure five key dimensions: meaning, positive relationships, engagement, positive emotions, and accomplishment. These dimensions were operationalized in an employee well-being questionnaire, as demonstrated in a study conducted by [120]. Example statements from the scale include: “My job has significance. In most cases, I can count on my colleagues. My job makes me happy.” and “My job inspires me. I look to the future with optimism. I will achieve what I want against all odds.” The Cronbach’s alpha is 0.897.

Employee innovative behavior: We applied the scales developed by [20] to measure EIB. The scale consists of six items, a sample item is “I often use new processes, techniques, and methods in my work.” The Cronbach’s alpha is 0.873.

We used SPSS 26 and SmartPLS 4.1.1.1 for the statistical analysis. PLS-SEM was selected because the model includes multiple latent constructs, indirect effects, and interaction effects, and because the study aims to estimate both the measurement and structural models simultaneously. PLS-SEM is also appropriate for examining complex predictive relationships with a moderate sample size [121]. The analysis proceeded in two stages. First, the measurement model was assessed for reliability and convergent and discriminant validity. Second, the structural model was evaluated by examining path coefficients, indirect effects, interaction effects, explanatory power, and collinearity. Mediation and moderation effects were tested using bootstrapping with 5,000 resamples, as illustrated in Figure 2Figure 2.

Our sample included 251 OBM professionals occupying diverse roles within product innovation teams. Table 1 summarizes demographic and professional characteristics. The Participants were 58.6% male and 41.4% female. Most participants were between the ages of 26 and 35, with 23.5% aged 26 - 30 and 25.5% aged 31 - 35. A majority held advanced degrees, including 41.8% with a Master’s, and participants reported a range of work experience, with most falling within the 1 - 4 years range.

Table 1. Sample demographics (n = 251).

Source(s): Created by authors.

To validate our research model, we examined reliability, convergent validity, and discriminant validity, with results presented in Tables 2-4. Reliability was assessed at both the item and construct levels. Item reliability was evaluated using outer loadings, all of which exceeded 0.5, with most above 0.7. These values indicate that the questionnaire items strongly reflect their intended constructs [122][123]. Construct reliability was further tested using Cronbach’s alpha (α) and composite reliability (CR). Both measures surpassed the accepted threshold of 0.70 [122][124], and in this study, both Cronbach’s alpha and CR values for all variables were above 0.80. This demonstrates strong internal consistency and supports the reliability of the measurement model. Convergent validity was also established. All factor loadings exceeded 0.5, and the average variance extracted (AVE) values for each construct were greater than 0.5 [125]. These findings confirm that each construct accounts for a meaningful proportion of variance in its indicators. We assessed discriminant validity using the Fornell-Larcker criterion [126], and the heterotrait-monotrait (HTMT) ratio [127]. These methods were tested to verify the correlation among latent variables. The HTMT is widely regarded as the most robust method for testing discriminant validity [122]. As shown in Table 3, all HTMT values ranged from 0.067 to 0.453, well below the recommended threshold of 0.85 [127]. Thereby confirming strong discriminant validity among the constructs. Additionally, Table 4 shows that the square root of the AVE for each construct is higher than its correlations with other variables. This confirms that the measurement model meets the standard for discriminant validity [127][128].

Table 2.Constructs reliability and (AVE) for reflective constructs.

Source(s): Created by authors.

Table 3.HTMT.

Note: Created by authors.

Table 4.Fornell and Larkers criterion.

Note: Created by authors.

Before examining the structural relationships hypothesized in this study, we examined the potential collinearity issues among constructs, in order to ensure accuracy and reliability in results interpretation. We used the Variance Inflation Factor (VIF) analysis to check for multicollinearity in the structural model. All VIF values ranged from 1.005 to 1.175, well below the threshold of 5 [122]. As shown in Table 5, the results confirm that multicollinearity is not a concern. This supports the model’s stability and reliability for testing the hypotheses. Following the collinearity assessment, we evaluated the path coefficients to test the proposed hypothesized relationships outlined in Figure 2Figure 2.

Table 5.Collinearity assessment (inner VIF values).

Note: Created by authors.

We tested the structural model using SmartPLS with a bootstrapping procedure of 5000 resamples. This approach evaluated the hypothesized relationships based on significance, path strength, and explanatory power [122][129]. Table 6 shows the path coefficients and their significance levels. The results show that five out of six proposed hypotheses are statistically supported. One hypothesis did not meet the significance criteria and was not supported. The model demonstrated acceptable explanatory power: the R² value for employee innovative behavior was 0.232 and for job security was 0.066, indicating that the predictors account for a meaningful share of variance in the outcome variables and are consistent with ranges reported in comparable behavioral PLS-SEM studies (Hair et al., 2019). The analysis began by testing the direct relationships in the conceptual framework. It first assessed the effect of AI adoption on employee innovative behavior to establish a baseline before exploring mediation and moderation effects.

Table 6.Results of the hypotheses testing.

Notes: ***p < 0.01; **p < 0.05; *p < 0.10.

First, we explored the direct relationships between variables. We found that AI adoption has a significant negative impact on employee innovative behavior (β = −0.232, t = 4.099, p < 0.01), which supports our H1. In examining further hypotheses, the impact of AI adoption on job security was assessed. AI adoption showed a significant negative effect on job security (β = −0.182, t = 2.678, p< 0.01), which supports H2a. Conversely, job security significantly and positively impacted employee innovative behavior (β = 0.394, t = 6.648, p < 0.01), which supports our H2b. Secondly, we calculated the indirect relationship between AI adoption and employee innovative behavior through job security. The mediation analysis shows that job security partially mediated the negative relationship between AI adoption and employee innovative behavior (β = −0.072, t = 2.713, p < 0.01). Thus, H2 was significantly supported. Thirdly, the moderated effects were explored by analyzing whether employee well-being could buffer or amplify the direct relationships examined earlier. Interestingly, the moderation analysis results revealed a non-significant moderation effect of employee well-being on the direct association between AI adoption and employee innovative behavior (β = −0.051, t = 0.914, p > 0.05), thus not supporting H3. The interaction between AI adoption and employee well-being was significantly associated with job security (β = 0.195, t = 3.278, p = 0.001). Conditional-effect analysis showed that when employee well-being was low (−1 SD), AI adoption was negatively associated with job security (β = −0.377, 95% CI [−0.509, −0.211]). At the mean level of employee well-being, the negative association remained significant (β = −0.182, 95% CI [−0.307, −0.039]). At a high level of employee well-being (+1 SD), the association became small and non-significant (β = 0.013, 95% CI [−0.176, 0.208]). These findings indicate that higher employee well-being substantially weakens the negative association between AI adoption and job security, consistent with H4.

Figure 2.Structural model.

We advance organizational behavior and technology-management scholarship by identifying the psychological pathway through which AI adoption influences innovative behavior in OBM firms, and by offering a theoretically integrated account of why that pathway operates. A recurrent critique of AI-and-work research is that TAM and COR do not individually address innovation. In this study, TAM is used as a contextual theoretical lens rather than as a fully tested explanatory model. Specifically, TAM helps explain why employees may respond differently to AI depending on whether they perceive it as useful and supportive in their work. However, perceived usefulness and perceived ease of use were not directly measured in the present study. Therefore, we do not test the core TAM mechanisms empirically. Instead, TAM provides a broader conceptual context for understanding employees’ responses to AI adoption, while COR theory directly explains the resource-related mechanisms involving job security, well-being, and innovative behavior.

This synthesis provides a more complete model than either theory offers independently. Prior TAM-based research emphasizes perceived usefulness and efficiency gains of AI [39][130], but our evidence shows that innovation declines when employees experience role uncertainty and resource threat [20][90][91]. Theoretically, we extend TAM by showing that perceived usefulness is not sufficient for positive behavioral outcomes when job-related concerns are salient. Integrating COR, we position job security as a central psychological resource; its erosion under AI implementation prompts withdrawal from discretionary, innovation-oriented behaviors. This expands COR’s application into technology-induced workplace transitions and emphasizes the need to address the emotional and psychological costs embedded in digital transformation, costs that efficiency-centered AI literature has largely set aside. It is worth noting, however, that job security perceptions are not shaped by AI alone. Macro-level factors, including economic volatility, industry restructuring, labor market tightness, and organizational culture, all influence baseline levels of perceived security. Our model captures the AI-specific contribution to these perceptions after controlling for individual characteristics, but future work should examine how contextual factors condition the AI-job security relationship. We also contribute by introducing employee well-being as a moderating factor. Although it does not buffer the direct effect of AI on innovation, it does reduce the negative impact of AI on job security, consistent with COR’s resource-buffer logic [102]. The lack of moderation at the innovation level suggests that resilience alone cannot offset structural concerns about role stability. Even so, well-being functions as a protective factor, enriching theorizing on psychological resources in disrupted work contexts [30]. Our research model, connecting AI adoption, job security, well-being, and innovative behavior, captures AI’s double-edged nature by shifting focus from system-level benefits to employee-level responses in innovation-intensive settings such as OBM. Finally, AI’s impact is contingent on how organizations manage employee perceptions during technological change. By combining TAM and COR and validating this integrated model in a high-stakes industry, we provide a lens for understanding the socio-psychological dynamics of AI in the workplace in OBM firms.

Our findings translate into four interconnected and directly evidence-based recommendations for OBM managers implementing AI. First, signal job security throughout the change process. Communicate the augmentation intent (not replacement), map how tasks will shift, and provide reskilling and internal-mobility pathways. These actions protect the psychological resource of job security and sustain discretionary innovation, consistent with COR and evidence that insecurity suppresses innovative effort. Second, design AI for augmentation using work-design principles, preserve autonomy, control, and skill variety in AI-enabled workflows and decision rights [50]. Adopt “collaborative intelligence” practices that keep people in the loop [100] and pursue human-machine complementarity rather than substitution [1]. These practices align with TAM by increasing perceived usefulness and adoption quality [39][130]. Third, invest in employee well-being as a buffer against AI-related threats. Provide mental-health resources, manageable workload norms, supportive leadership, and flexible work where feasible, interventions that enlarge resource reserves and mitigate insecurity [93][102][103]. Our results show that well-being attenuates AI’s negative effect on job security, creating the conditions for innovation to thrive. Fourth, establish structured AI onboarding and change management: AI literacy training, safe sandboxes for experimentation, transparent decision-logic briefings, and rapid feedback channels. These steps enhance perceived usefulness and ease of use (core TAM constructs) TAM and reduce threat perceptions [24][131]. Across all four recommendations, firms should additionally commit to transparent, human-centered AI governance: communicating clearly how AI decisions are made, conducting fairness checks, and maintaining proportionate oversight that protects both performance goals and employee dignity [132]-[134]. Together, these practices enable firms to realize AI’s efficiency gains without eroding the climate for employee innovation.

We acknowledge that this research has limitations that need to be addressed in future studies. The first limitation is that our model considered only two psychological variables: job security and well-being. Future research should consider additional factors, such as leadership style, team support, and trust in AI, that may shape these outcomes. Moreover, the indirect effect of well-being suggests more complex underlying mechanisms (e.g., stress coping, organizational trust), which longitudinal or multi-level designs could further uncover. Second, this study does not account for macro-level factors such as broader economic conditions, organizational culture, or national labor-market context, all of which may independently shape perceived job security. Future research should examine whether the AI-job security relationship is moderated by organizational climate (e.g., psychological safety) or labor-market tightness, which may strengthen or attenuate the effects observed here. Third, while the current PLS-SEM model demonstrates strong reliability and validity, future studies would benefit from complementary robustness checks such as alternative model specifications, holdout sample validation, or the inclusion of additional control variables (e.g., firm AI maturity, tenure) to further confirm the stability of the structural estimates. Finally, our study focuses on the OBM, where innovation is a core value. The findings may not generalize to low-innovation sectors or jobs with routine tasks. Replication in other contexts, such as healthcare, logistics, or education, could validate and refine the model’s boundaries.

Despite these limitations, the study makes significant contributions by bridging psychological insights with the predominantly technical discourse on AI. It provides both practical and theoretical frameworks that are valuable for academics and professionals. To strengthen and generalize these findings, future studies should explore new variables, apply the framework across diverse industries, and examine how cultural context shapes the AI-employee dynamic. These efforts will contribute to a deeper understanding of how AI affects human behavior and organizational dynamics. This research enhances our understanding of the psychological processes that link AI adoption to employee innovation, and identifies meaningful directions for future inquiry.