Neural hijacking or empowerment? A conceptual framework on how predictive HR algorithms shape employee motivation and workforce sustainability

The standard defense of predictive HR algorithms is efficiency: they process more data, more consistently, than any human manager can. What this defense misses is that efficiency in behavioral influence is not a neutral property. Organizations now routinely use predictive algorithms to screen candidates, flag attrition risk, personalize learning content, automate performance feedback and map career trajectories (Jarrahi et al., 2021; Kellogg et al., 2020). The efficiency argument is simple: these systems process behavioral and performance data at a scale no human analyst can match, and they do so continuously (Huang and Rust, 2018; Raisch and Krakowski, 2021). The gains in speed and coverage are real. What is considerably less understood is what these systems do to the people operating inside them.

The problem is not administrative. When predictive HR algorithms move beyond processing data and begin shaping behavior through gamified progress trackers, streak-based nudges, micro-reward notifications and engagement-optimized feedback cycles, they start operating on the psychological mechanisms that govern motivation and learning (Fogg, 2003; Eyal, 2014). These features are not coincidental design choices. They are borrowed directly from gaming and social media, where the same techniques were developed and refined specifically to sustain platform use (Montag et al., 2019; Alter, 2018; Deterding et al., 2011). They work well for that purpose. Whether sustained platform engagement and genuine skill development point in the same direction is a separate question, and the evidence suggests they frequently do not (Deci et al., 1999; Mekler et al., 2017; Gneezy et al., 2011).

This paper introduces Neural Hijacking as a conceptual construct to name a specific risk embedded in this design logic. When predictive HR algorithms optimize for engagement metrics rather than developmental outcomes, they activate dopaminergic reward pathways through frequent micro-rewards and behavioral nudges (Schultz et al., 1997; Volkow et al., 2011), and over time, they recalibrate employees’ motivational orientation away from self-directed learning and toward externally cued compliance. What makes this distinct from conventional forms of managerial control is that it operates through system design rather than explicit instruction, and most employees do not recognize it as influence at all (Eslami et al., 2015; Susser et al., 2019; Burrell, 2016).

Neural Hijacking also differs from the long-established undermining effect of extrinsic rewards (Deci et al., 1999; Frey and Jegen, 2001). That literature documents how external incentives in general can crowd out intrinsic motivation. Neural Hijacking is more specific: it describes what happens when the incentive structure is algorithmically adaptive, personalized continuously from behavioral data and optimized for retention rather than growth. Yeung (2017) describes this design logic as hypernudging, where data-driven systems invisibly reconfigure choice environments in real time. Applied to HR, this creates a risk that has not been adequately theorized: employees who appear engaged by every available organizational metric may be drifting away from the kind of self-directed, motivated learning that sustains workforce adaptability over time (Zuboff, 2019).

The practical problem is more specific than it first appears. Organizations are not choosing between AI and no AI. They are choosing, usually without realizing it, between algorithmic systems that build motivational capacity and systems that deplete it while reporting engagement metrics that make the depletion invisible. The framework developed here is designed to make that choice legible. Under what conditions do predictive HR algorithms support employee empowerment, building genuine autonomy, competence and adaptability? And under what conditions do they activate a hijacking pathway, one that produces surface engagement and behavioral dependency while depleting the motivational foundations that organizations actually need for long-run performance? These questions matter practically. Organizations that drive engagement metrics upward while quietly eroding intrinsic motivation face a compounding problem: the workforce looks productive by conventional measures while its capacity for creativity, adaptation and self-initiated growth declines (Kellogg et al., 2020; Raisch and Krakowski, 2021).

To address these questions, this paper develops a dual-pathway conceptual framework grounded in Self-Determination Theory (SDT) (Deci and Ryan, 1985; Ryan and Deci, 2000), Nudge Theory (Thaler and Sunstein, 2008; Sunstein, 2016) and neuroscientific research on reward processing and motivational recalibration (Schultz et al., 1997; Robinson and Berridge, 2008; Volkow et al., 2011). The framework maps the design antecedents, psychological mechanisms and workforce-level outcomes associated with each pathway and translates these into five proposition sets for empirical investigation. The paper concludes with implications for HR practitioners, AI system designers and policymakers navigating the governance of algorithmic management systems. This paper departs from existing work by rejecting the assumption that motivational dynamics in organizations operate within stable, human-mediated environments. In algorithmically adaptive contexts, this assumption no longer holds.

Predictive HR algorithms use machine learning, natural language processing and behavioral data analytics to generate probabilistic assessments of workforce dynamics. In practice, this means systems that flag employees at attrition risk before managers notice any behavioral signal, recommend personalized learning sequences that adapt in real time to a learner’s pace and accuracy, evaluate candidate pools at scale and generate automated performance feedback (Jarrahi et al., 2021; Kellogg et al., 2020). The operational scope of these systems has expanded considerably. What began as back-office analytics has become a continuous, ambient layer of influence over how employees experience work.

Two structural risks accompany this expansion. The first is opacity. Most predictive HR algorithms function as black boxes: neither employees nor their managers can inspect the logic behind a recommendation, a risk score or a feedback rating (Burrell, 2016). Eslami et al. (2015) showed that users of algorithmic systems routinely construct inaccurate causal explanations for outputs they cannot see, which means employees may be responding to algorithmic signals they fundamentally misunderstand. The second risk is behavioral drift. As these systems optimize iteratively against engagement metrics, they tend to reward activity over outcome, reinforcing behaviors that generate data rather than behaviors that indicate genuine learning or development (Shin, 2020; Berman and Katona, 2020). Over time, the metric becomes the target, and the developmental goal it was meant to approximate gets quietly displaced.

Together, these risks suggest that predictive HR algorithms are not neutral administrative tools. They function as behavioral architects, shaping what employees attend to, what they pursue and how they understand their own performance.

The concept of algorithmic management describes the delegation of managerial functions, task allocation, performance monitoring and behavioral direction to automated systems (Kellogg et al., 2020). Research in this area has documented a consistent tension between algorithmic efficiency and worker autonomy. Lee et al. (2015) found that workers managed by algorithmic systems reported reduced sense of control over their work, even when the systems were designed to support rather than restrict them. Möhlmann and Zalmanson (2017), studying platform workers, found that algorithmic direction eroded perceived autonomy even among workers who formally retained the right to refuse tasks.

This matters for HR systems because autonomy is not simply a preference. Within SDT (Deci and Ryan, 1985; Ryan and Deci, 2000), it is a fundamental psychological need whose satisfaction is directly linked to intrinsic motivation, sustained engagement, and well-being. When algorithmic management structures narrow the range of choices available to employees, or when it steers behavior so consistently that choice feels nominal, the psychological conditions for intrinsic motivation begin to deteriorate. Raisch and Krakowski (2021) frame this as the automation–augmentation paradox: the same AI systems that enhance organizational capability can simultaneously reduce the human capacity that organizational capability ultimately depends on.

What is undertheorized in this literature is the mechanism. Algorithmic management research documents the outcomes of constrained autonomy but says less about how specific design features produce those outcomes at the level of individual motivation. That is the gap this paper addresses.

Nudge Theory holds that the architecture of choices, how options are presented, sequenced and defaulted, shapes decisions without restricting them (Thaler and Sunstein, 2008). In AI-powered HR systems, nudging takes the form of push notifications timed to behavioral patterns, default enrollment in recommended learning modules, progress bars calibrated to sustain completion momentum and leaderboard rankings designed to activate social comparison (Sunstein, 2016). These features are not ethically neutral. Sunstein (2016) distinguishes between nudges that expand welfare by helping people act on their own stated preferences and nudges that serve the system’s objectives at the expense of the user’s. The distinction is consequential. In HR systems optimized for platform engagement rather than employee development, the nudge infrastructure often serves organizational metrics first.

Gamification intensifies this dynamic. Deterding et al. (2011) define gamification as the application of game design elements in non-game contexts, and in corporate learning platforms, this typically means points, badges, streaks and leaderboard rankings layered onto training content. The motivation research is instructive here. Mekler et al. (2017) found that individual gamification elements have differentiated effects: points and leaderboards consistently undermined intrinsic motivation, while progress indicators had more neutral or positive effects depending on implementation. Koivisto and Hamari's (2019) review of the gamification literature found that most studies report short-term engagement gains but far fewer demonstrate durable learning or behavioral transfer. Seaborn and Fels (2015) reach a similar conclusion: gamification works for compliance, less reliably for competence.

The mechanism that connects these design features to motivational harm runs through what Fogg (2003) called persuasive technology and what Eyal (2014) later formalized as the hook model: trigger, action, variable reward and investment. Variable reward, specifically the unpredictable delivery of positive feedback, is the design feature most directly linked to compulsive engagement in social media and gaming contexts (Alter, 2018; Montag et al., 2019). When HR platforms replicate this logic, as many now do through adaptive feedback timing and algorithmically varied encouragement, they import not just the engagement effects but the dependency risks as well. Susser et al. (2019) argue that influence crosses into manipulation when it bypasses rational agency. Algorithmic systems that use behavioral data to identify and exploit individual psychological vulnerabilities in the service of engagement metrics cross that boundary by design. Yeung (2017) calls this hypernudging: continuous, invisible reconfiguration of the choice environment based on real-time data. In an HR context, this means the system is not presenting a learning opportunity. It is engineering the conditions under which an employee will feel compelled to take it.

The neurological foundation for the neural hijacking argument rests on dopaminergic reward processing. Schultz et al. (1997) demonstrated that dopamine neurons in the midbrain do not simply respond to rewards; they respond to reward prediction errors, the difference between expected and received outcomes. Systems that deliver frequent, variable micro-rewards exploit this mechanism directly, generating repeated dopamine responses that reinforce the behavior preceding the reward. This is how habits form, and it is also how compulsive engagement patterns develop (Robinson and Berridge, 2008).

Volkow et al. (2011) argue that addiction-related changes in motivation extend beyond dopamine reward circuitry to affect the prefrontal systems involved in executive control and self-regulation. Prolonged exposure to high-frequency reward delivery weakens the capacity for delayed gratification, long-range planning and intrinsically motivated goal pursuit. Applied to HR systems, the implication is direct: employees whose learning experience is dominated by badge notifications, streak alerts and leaderboard movements may be training their reward systems to expect and require those signals, while the neural architecture supporting deep, self-directed learning is correspondingly underexercised.

This is also consistent with the motivation crowding literature in economics. Frey and Jegen (2001) showed that external interventions undermine intrinsic motivation when they are experienced as controlling rather than supportive. Gneezy et al. (2011) extended this by demonstrating that the crowding effect is conditional: incentives undermine intrinsic motivation most reliably when the activity has inherent interest and when the external reward is perceived as the reason for engaging. Algorithmically delivered micro-rewards in professional development contexts meet both conditions. The activity, skill development, has inherent value to employees. And the reward structure, visible, frequent and tied to measurable behaviors, is explicitly framed as the reason to engage.

Workforce sustainability describes an organization’s capacity to maintain and develop a workforce that is capable, motivated and adaptive over time, not just productive in the near term. It encompasses employee well-being, psychological safety, skill renewal and the organizational conditions that allow intrinsic motivation to function (Kramar, 2014). An HR strategy that optimizes exclusively for engagement metrics risks producing a workforce that performs well on those metrics while gradually losing the deeper capabilities that organizational performance ultimately depends on: creative problem-solving, collaborative judgment and self-initiated learning.

The risk compounds over time. If algorithmic HR systems recalibrate employees’ motivational orientation toward externally cued compliance, as the neural hijacking argument suggests, the workforce becomes progressively less capable of the kind of self-directed, adaptive behavior that unpredictable environments require. Zuboff (2019) frames this in broader terms: systems designed to predict and modify behavior at scale produce populations that are increasingly legible to the system and decreasingly capable of autonomous action. The organizational HR system is a microcosm of exactly that dynamic.

The synthesis across these four literature points to a specific conceptual need. The algorithmic management literature documents constrained autonomy but does not theorize the motivational mechanism. The gamification and nudging literature identifies undermining effects but does not connect them to a neurological model. The neuroscience literature provides the mechanism but has not been applied to HR system design. And the workforce sustainability literature identifies the long-term stakes without specifying the causal pathway from design to outcome. The Neural Hijacking framework proposed in this paper addresses that gap by integrating these four bodies into a single explanatory structure.

Each of the theoretical bodies reviewed above makes a genuine contribution to understanding algorithmic influence on motivation. None of them, individually or in combination, is sufficient to explain the specific risk this paper addresses. Making that insufficiency explicit is necessary before the Neural Hijacking construct can be positioned as a theoretical contribution rather than a relabeling exercise.

SDT is the most established motivational framework in organizational psychology, and its predictions about extrinsic reward undermining are empirically well-supported (Deci et al., 1999; Ryan and Deci, 2000). But SDT was developed to explain motivation in social environments where the agents of need support or thwarting are human: managers, peers and institutional structures that employees can negotiate with, contest and occasionally change. SDT does not theorize what happens when the environmental agent is an adaptive algorithm that learns from behavioral data, self-optimizes in real time and operates invisibly to the people it influences. More specifically, SDT’s undermining effect predicts that extrinsic rewards reduce intrinsic motivation, but it does not specify the neurological mechanism through which this occurs, nor does it account for the possibility that the reward structure itself is personalized and continuously recalibrated to each individual’s behavioral vulnerabilities. An algorithm that learns which notification timing, reward type and social comparison frame maximize a specific employee’s compliance is not simply delivering extrinsic rewards. It is engineering the conditions under which extrinsic regulation feels, to that employee, like autonomous choice. SDT has no theoretical language for this distinction.

Nudge Theory addresses choice architecture, and Yeung’s (2017) hypernudging concept extends its usefully into data-driven contexts. But Nudge Theory, even in its extended form, remains agnostic about motivation. It explains how behavior is steered but says nothing about what that steering does to the psychological structures that govern future motivation. A person who is repeatedly nudged into a behavior may comply without any motivational consequence or may develop the kind of dependency that Robinson and Berridge (2008) associate with incentive sensitization. Nudge Theory cannot distinguish between these outcomes because it does not engage with motivational architecture at the level of psychological need or neurological mechanism.

The algorithmic management literature documents that workers under algorithmic direction experience reduced autonomy and increased surveillance (Kellogg et al., 2020; Lee et al., 2015; Möhlmann and Zalmanson, 2017). This is an important empirical finding, but it describes an outcome without specifying the mechanism that produces it, and it treats constrained autonomy as a relatively uniform experience. The literature does not theorize why some employees under identical algorithmic management conditions maintain intrinsic motivation while others experience rapid erosion, nor does it connect the design features of algorithmic systems to the neurological processes that govern motivational orientation.

The motivation crowding literature in economics (Frey and Jegen, 2001; Gneezy et al., 2011) comes closest to the Neural Hijacking argument by showing that external interventions crowd out intrinsic motivation under specific conditions. But crowding theory was developed in contexts of fixed incentive structures, such as payment for blood donation or fines for late school pickup. It does not account for algorithmically adaptive incentive structures that learn from individual behavioral data, personalize reward delivery and self-optimize for engagement in ways that users cannot detect or contest.

Neural Hijacking is needed precisely because no existing construct theorizes the intersection of these four failures: the invisible, personalized, neurologically targeted and continuously optimizing nature of algorithmic reward delivery in HR contexts. The construct is not SDT plus nudging plus neuroscience. It is the specific phenomenon that emerges at the intersection of all three and that none of them, alone or combined, can fully account for.

SDT (Deci and Ryan, 1985; Ryan and Deci, 2000) holds that sustained, high-quality motivation depends on the satisfaction of three basic psychological needs: autonomy, the experience of volition and self-endorsed choice; competence, the perception of genuine mastery and effectiveness; and relatedness, the sense of meaningful connection with others. When these needs are met, people tend toward intrinsic motivation, engaging in activities because they find them inherently interesting or personally meaningful. When these needs are thwarted, motivation shifts toward extrinsic regulation or amotivation, states associated with lower performance quality, reduced creativity and higher burnout risk (Ryan and Deci, 2000; Deci et al., 1999).

The theory has direct implications for how predictive HR algorithms are designed. An algorithm that surfaces personalized development opportunities aligned with an employee’s own career goals, adjusts challenge level to sustain a sense of genuine progress and facilitates meaningful collaboration supports all three needs in ways that reinforce intrinsic motivation. The same algorithm, redesigned to maximize platform engagement through badge systems, streak counters and leaderboard rankings, tells a different story. Autonomy is nominally preserved but practically narrowed: the employee chooses from options the algorithm has curated. Competence is signaled not through genuine mastery but through task completion metrics. Relatedness is reduced to social comparison scores (Ryan and Deci, 2024). SDT predicts, and Deci et al.’ s (1999) meta-analysis confirms, that this shift from need-supportive to need-thwarting design reliably undermines intrinsic motivation over time.

What SDT does not fully address is the mechanism through which algorithmically delivered rewards operate on motivational systems neurologically. That requires a different theoretical register, addressed in Section 3.3.

Nudge Theory argues that the design of choice environments, how options are framed, sequenced, defaulted and presented, shapes decisions in predictable ways without restricting freedom of choice (Thaler and Sunstein, 2008). In principle, nudging respects autonomy: people can always opt out, and the nudge is meant to make it easier for them to act on what they already value. In practice, the line between nudging and behavioral engineering is less stable than the theory implies.

Sunstein (2016) distinguishes between what he calls welfare-promoting nudges, which help people act consistently with their own goals, and nudges that serve the system designer’s objectives at the user’s expense. This distinction is critical for HR systems. A nudge that reminds an employee about a skill module she flagged as relevant to her development goals is welfare-promoting. An algorithm that sends a push notification at the moment she is statistically most likely to click, based on her behavioral history, regardless of whether the content serves her development, is something closer to what Susser et al. (2019) call online manipulation: influence that bypasses rational agency by exploiting psychological vulnerabilities rather than addressing stated preferences.

Yeung (2017) extends this analysis with the concept of hypernudging, describing how big data systems continuously reconfigure choice environments in real time, making the nudge invisible, personalized and effectively inescapable. Standard nudge theory assumes a relatively static choice architecture and a passive choice architect. Hypernudging involves an architecture that adapts itself to each individual user, learns which stimuli generate compliance and self-optimizes accordingly. When HR platforms use A/B testing, adaptive feedback timing and machine learning to identify which notification format, reward type and social comparison frame maximizes individual engagement, they are not nudging in Thaler and Sunstein’s original sense. They are running a continuous optimization experiment on their own users (Fogg, 2003; Eyal, 2014).

The tension with SDT is worth naming directly. SDT holds that autonomy support is essential for intrinsic motivation. Nudge theory, particularly in its hypernudging form, involves structuring the environment so that autonomous choice produces the outcomes the system designer wants. These two frameworks are not simply complementary. They are in partial conflict, and the Neural Hijacking construct sits precisely at that conflict point. When algorithmic design uses nudging techniques to produce behaviors that look like autonomous engagement but are in fact environmentally engineered, the SDT requirement for genuine autonomy support is violated even though no explicit mandate has been issued.

The neurological foundation for Neural Hijacking rests on dopaminergic reward processing. Schultz et al. (1997) demonstrated that midbrain dopamine neurons respond not to rewards themselves but to reward prediction errors: the gap between expected and received outcomes. When an outcome is better than predicted, dopamine release spikes, reinforcing the preceding behavior. When an outcome matches prediction, the dopamine response is flat. This mechanism is why variable reward delivery, the kind built into slot machines, social media feeds and increasingly into HR platform feedback systems, is so effective at sustaining behavioral engagement. The reward is unpredictable enough to keep prediction errors positive and frequent enough to generate repeated reinforcement.

Robinson and Berridge (2008) argue that dopaminergic systems govern wanting rather than liking. Repeated activation of these systems through frequent reward delivery increases the motivational salience of reward-associated cues, making employees more responsive to the platform’s signals over time, regardless of whether those signals correspond to genuine developmental value. Volkow et al. (2011) extend this further, arguing that sustained high-frequency reward exposure affects prefrontal systems governing executive control and self-regulation. The capacity for delayed gratification, for sustained effort on tasks without immediate feedback and for self-directed goal pursuit weakens under conditions of prolonged reward dependency. In an HR learning context, this means employees may find it progressively harder to engage with complex, slow-burn developmental work precisely because their reward systems have been conditioned to expect and require constant algorithmic feedback.

Alter (2018) and Montag et al. (2019) document how these same neurological mechanisms underlie addictive engagement patterns in social media and gaming environments. The HR platform is a different context but not a mechanistically different system. When it imports the design logic of variable reward delivery, social comparison triggers and streak-based loss aversion, it imports the associated psychological risks as well.

The empowerment pathway, by contrast, engages reward systems in a way that sustains cognitive effort rather than bypassing it. Challenges calibrated to be demanding but achievable activate dopaminergic reward responses tied to genuine mastery, the kind of response Schultz et al. (1997) associate with learning rather than habit formation. Meaningful feedback that conveys progress toward self-selected goals supports the competence need in SDT terms while also generating reward responses that reinforce continued engagement with the developmental task rather than with the platform.

The three theoretical bodies reviewed above are not simply parallel perspectives on the same phenomenon. They operate at different levels of analysis and explain different parts of the same causal chain.

SDT operates at the level of psychological need satisfaction and explains which design features will support or undermine the motivational conditions for sustained, high-quality engagement. Nudge Theory operates at the level of choice architecture and explains the mechanism by which algorithmic systems steer behavior without explicit instruction. Neuroscience operates at the level of neural mechanisms and explains why certain design features, specifically variable reward delivery and high-frequency micro-rewards, produce the motivational recalibration that Neural Hijacking describes.

The integration of the three produces a coherent explanatory structure. HR algorithms designed with ethical AI principles and employee goal alignment support SDT need satisfaction, use welfare-promoting nudges and activate reward systems in ways that sustain cognitive engagement and genuine skill development. This is the empowerment pathway. HR algorithms optimized for engagement metrics and platform retention thwart SDT need satisfaction through narrowed choice and superficial competence signals, use hypernudging techniques that exploit behavioral vulnerabilities and activate reward systems in ways that build dependency and erode the capacity for self-directed learning. This is the neural hijacking pathway.

The theoretical contribution of this integration is not additive but constitutive. SDT identifies which psychological conditions matter. Nudge Theory identifies the design mechanism that affects those conditions. Neuroscience identifies the biological process through which design features produce durable motivational change. Neural Hijacking names the phenomenon that only becomes visible when all three levels of analysis are held simultaneously: the process by which algorithmically engineered, personalized, neurologically targeted reward delivery recalibrates employees’ motivational orientation in ways that are invisible to the individuals experiencing them, inconsistent with their developmental interests and structurally incentivized by the vendor relationships governing most HR platform markets. This is what existing theory cannot explain. This is what the framework in Section 4 is built to address.

The framework developed in this paper positions predictive HR algorithms as a behavioral mediator whose effects on employee motivation and workforce sustainability depend on three interacting layers: the design antecedents that shape how a system is built and deployed, the psychological mechanisms through which the system operates on individual motivation, and the organizational and individual moderators that determine which pathway a given implementation activates. Two distinct pathways emerge from this structure. The empowerment pathway describes algorithmic systems that support autonomy, genuine competence development and relatedness, producing sustained intrinsic motivation and long-term workforce adaptability. The neural hijacking pathway describes systems optimized for engagement metrics and platform retention, producing surface-level compliance, motivational dependency and progressive skill stagnation.

These pathways are not fixed categories. A single HR system can shift between them depending on organizational pressures, algorithmic updates and changes in the performance metrics that govern system optimization. The framework is therefore better understood as a continuum than a binary, with empowerment and neural hijacking as poles rather than pigeonholes. This is discussed further in Section 4.5.

The antecedents are the foundational conditions that make one pathway more likely than the other. They operate at two levels: organizational intent and system design ethics.

At the organizational level, the primary antecedent is the design objective embedded in the HR system’s optimization target. Organizations that define algorithmic success in terms of daily active users, module completion rates or platform engagement time are structurally biasing their systems toward the neural hijacking pathway, because those metrics reward the design features, variable rewards, streak mechanics and social comparison triggers that produce engagement dependency rather than developmental depth (Kellogg et al., 2020; Shin, 2020). Organizations that define success in terms of skill transfer, career progression or employee-reported growth align their optimization targets with the empowerment pathway.

At the design level, the critical antecedents are transparency, explainability and data governance. Burrell (2016) argues that opacity in algorithmic systems is not simply a technical limitation but a design choice with ethical consequences: systems that cannot explain their recommendations to users cannot support informed autonomy, even when they formally preserve freedom of choice. Employees who cannot understand why an algorithm is recommending a particular learning path, flagging them as an attrition risk or rating their performance at a particular level are not in a position to exercise meaningful agency over their own development (Eslami et al., 2015). Informed consent over data use and genuine employee agency in algorithmic decision-making are antecedents to the empowerment pathway. Their absence is an antecedent to hijacking.

Ethical AI design principles, including fairness, bias mitigation and human oversight, function as additional antecedents (Jarrahi et al., 2021; Raisch and Krakowski, 2021). Systems built without bias auditing may systematically disadvantage particular employee groups, compounding motivational harm with structural inequity. The empowerment pathway requires these principles to be embedded at the design stage, not retrofitted after deployment.

The mechanisms are the specific algorithmic features through which antecedent conditions translate into psychological outcomes. They operate differently across the two pathways.

On the empowerment pathway, three mechanisms are central. Challenge calibration describes the process of matching task difficulty to employee capability in ways that sustain genuine effort without producing overwhelm or boredom, conditions that Deci and Ryan (1985) associate with competence need satisfaction and that are associated with deeper cognitive processing and knowledge transfer. Growth-oriented feedback describes qualitative, developmental information that helps employees understand what they are learning and why it matters, rather than numerical scores that signal performance without supporting understanding (Ryan and Deci, 2000). Personalized development pathways describe adaptive recommendations built around employee-stated career goals and interests, a design feature that supports autonomy by giving employees genuine input into the direction of their development (Möhlmann and Zalmanson, 2017).

On the neural hijacking pathway, three corresponding mechanisms operate. Choice architecture narrowing describes the gradual reduction of behavioral options through default settings, curated recommendation feeds and nudge sequences that guide employees toward algorithmically preferred actions while maintaining the appearance of free choice (Thaler and Sunstein, 2008; Yeung, 2017). Engagement-only optimization describes the system-level prioritization of usage metrics over developmental outcomes, producing algorithmic recommendations that maximize time-on-platform without regard to whether that time produces genuine skill development (Koivisto and Hamari, 2019; Seaborn and Fels, 2015). Gamified micro-reward cycling describes the use of badges, streaks, points and leaderboard rankings to deliver frequent, variable rewards that activate dopaminergic reinforcement loops, building behavioral dependency on platform signals rather than intrinsic engagement with learning content (Schultz et al., 1997; Mekler et al., 2017; Robinson and Berridge, 2008).

A framework that maps antecedents to mechanisms to outcomes without acknowledging the conditions that moderate these relationships would be incomplete. Three moderating factors are theoretically significant.

Employee algorithmic literacy describes the degree to which employees understand how the systems managing their development actually work. Eslami et al. (2015) found that users with lower algorithmic awareness were more susceptible to algorithmic influence and less able to critically evaluate system outputs. Employees with higher algorithmic literacy are better positioned to recognize nudging, question recommendations and maintain self-directed agency within algorithmically managed environments. This suggests that the neural hijacking pathway is moderated by awareness: the same system design may produce dependency in employees with low algorithmic literacy and critical engagement in those with high literacy.

Organizational culture moderates the relationship between design antecedents and pathway activation. An organization that explicitly values employee development, supports managerial override of algorithmic recommendations and treats engagement metrics as one input among many rather than the primary performance indicator creates conditions that buffer against hijacking even when the underlying system has hijacking-prone features (Kellogg et al., 2020; Raisch and Krakowski, 2021). Conversely, a culture that treats algorithmic outputs as authoritative and optimizes managerial incentives around engagement dashboards amplifies the hijacking pathway regardless of the system’s stated design intent.

Individual motivational orientation, specifically the degree to which an employee begins with high versus low intrinsic motivation in a given domain, moderates the speed and severity of motivational recalibration. Frey and Jegen (2001) showed that motivation crowding effects are strongest when intrinsic motivation is initially high. Employees who enter an algorithmically managed learning environment with strong intrinsic interest in the domain may experience faster motivational erosion under neural hijacking conditions than those who were already extrinsically oriented, because the displacement effect operates on something real and present.

Figure 1 presents the dual-pathway model as a dynamic structure rather than a static binary. The diagram maps the causal flow from design antecedents through algorithmic mechanisms, moderated by organizational and individual conditions, to psychological and workforce-level outcomes. Feedback arrows from outcomes back to antecedents capture the dynamic possibility described in Section 4.6: that outcome pressures such as declining engagement metrics or increasing attrition can trigger organizational responses that push algorithm design in either direction.

The left pathway traces ethical AI design and employee-centered optimization through challenge calibration, growth-oriented feedback and personalized development to outcomes of sustained intrinsic motivation, deep learning and long-term adaptability. The right pathway traces engagement-metric optimization and opaque design through choice narrowing, gamified micro-reward cycling and engagement-only feedback to outcomes of surface compliance, skill stagnation and burnout risk. The moderating layer sits between mechanisms and outcomes, capturing how algorithmic literacy, organizational culture and individual motivational orientation shape which outcomes a given implementation produces.

The empowerment and neural hijacking pathways are analytical constructs, not descriptions of fixed system types. In practice, most HR systems sit somewhere between the poles, and many shift position over time. An algorithm initially designed around employee development goals may drift toward the hijacking end when organizational leadership prioritizes engagement metrics under competitive pressure, when a platform update introduces gamified features to improve retention numbers, or when the vendor’s optimization model is updated without transparency to the client organization (Zuboff, 2019; Kellogg et al., 2020).

This drift is rarely deliberate. It tends to happen incrementally, through a series of individually defensible design decisions, each one optimizing for a measurable metric, that collectively recalibrate the system’s behavioral architecture away from development and toward engagement dependency. Burrell (2016) notes that algorithmic systems often produce outcomes their designers did not intend and cannot fully explain, precisely because the optimization process operates on proxies rather than on the underlying goals those proxies are meant to represent.

The continuum perspective has a practical implication. Organizations cannot assess their HR systems once at deployment and assume the assessment holds. The pathway a system activates is a function of its current design, its current optimization targets and its current organizational context, all of which can change. Regular calibration audits, in which engagement metrics are examined alongside developmental outcome measures, are the practical mechanism for tracking position on the continuum and detecting drift before its motivational consequences become entrenched.

The framework developed in Sections 3 and 4 generates a set of theoretically grounded propositions linking design antecedents, psychological mechanisms, moderating conditions and workforce outcomes across the two algorithmic pathways. These propositions are intended to be specific enough to operationalize, grounded in prior empirical findings and non-tautological: each one makes a claim about a relationship that could, in principle, be disconfirmed.

The antecedent layer of the framework holds that the organizational objectives embedded in system design are the primary determinant of which pathway a predictive HR algorithm activates. This is not simply a definitional claim. Kellogg et al. (2020) document that algorithms optimized for task throughput produce worker experiences characterized by reduced autonomy and increased surveillance, even when designers intended otherwise. Shin (2020) shows that recommender system design choices directly shape user behavior in ways that users cannot detect. The following propositions test whether these effects hold in HR-specific contexts and whether ethical design principles buffer against them.

The distinction between P1a and P1b matters. P1a tests the explainability mechanism at the individual level. P1b tests the optimization target effect at the system level over time. P1c introduces data agency as a moderator, an element absent from the original propositions but theoretically necessary given the framework’s antecedent structure (Burrell, 2016; Eslami et al., 2015).

The empowerment mechanism propositions test whether specific design features produce the psychological outcomes the framework predicts. The empirical anchoring for these propositions comes from SDT research (Deci et al., 1999; Ryan and Deci, 2000) and from the gamification literature’s findings on differential element effects (Mekler et al., 2017).

P2b and P2c together test the competence-need mechanism at different points in the learning process: feedback quality affects what employees take away from completed tasks; challenge calibration affects engagement during the task itself. Separating these tests allows future researchers to identify which mechanism contributes more to developmental outcomes under different implementation conditions.

These propositions test the specific design features hypothesized to produce motivational harm. The empirical base here draws on the undermining effect literature (Deci et al., 1999; Frey and Jegen, 2001; Gneezy et al., 2011), gamification research (Koivisto and Hamari, 2019; Seaborn and Fels, 2015) and reward processing research (Schultz et al., 1997; Robinson and Berridge, 2008).

P3a separates the short-term engagement effect from the long-term retention effect, which the original proposition conflated. P3b operationalizes behavioral dependency in a way that can be measured. P3c introduces a measurable operationalization of choice narrowing that can be tracked in platform data.

These propositions test the moderating layer developed in Section 4.4, which was entirely absent from the original framework’s propositions. Without moderator propositions, the framework cannot explain why identical system designs produce different outcomes across employees and organizations, a question any empirical reviewer will raise.

The continuum perspective developed in Section 4.6 generates propositions about how systems change over time rather than simply about their static effects. This is theoretically important because it moves the framework beyond a snapshot model toward a dynamic account of algorithmic influence.

These five proposition sets establish a structured empirical agenda. They can be tested through longitudinal field studies tracking employees across algorithmic system updates, controlled experiments comparing platform design variants, experience sampling methods capturing motivational states at multiple points during algorithmic interactions and neuroimaging studies examining reward system activation under different feedback design conditions. Together, they position the Neural Hijacking construct as an empirically accessible rather than merely rhetorical contribution to the HR analytics literature.

SDT was developed to explain motivation in contexts where the social environment, managers, peers, institutional structures, supports or thwarts psychological need satisfaction (Deci and Ryan, 1985; Ryan and Deci, 2000). The framework developed here extends SDT into a context its originators did not anticipate: one where the primary environmental actor is not a person but an adaptive algorithmic system that learns from behavioral data, adjusts its outputs in real time and optimizes for objectives that may or may not align with the employee’s own developmental needs. This extension is not trivial. In human-managed environments, need thwarting tends to be visible, contestable and at least partially responsive to employee pushback. In algorithmically managed environments, need thwarting operates through design features that most employees cannot see, do not recognize as influence, and have limited ability to contest (Eslami et al., 2015; Burrell, 2016). The Neural Hijacking construct identifies this gap and gives SDT researchers a theoretically grounded entry point into algorithmic management contexts.

The framework also contributes to SDT by specifying a neurological mechanism through which need thwarting operates under algorithmic conditions. Deci et al.’,s (1999) meta-analysis established that extrinsic rewards undermine intrinsic motivation, but the mechanism remained largely psychological. By integrating Schultz et al.’,s (1997) reward prediction error model and Volkow et al.’ s (2011) work on reward dependency, the framework offers a neurological account of why algorithmically delivered micro-rewards produce motivational erosion: they recalibrate the dopaminergic systems that govern motivational orientation, not just the cognitive appraisals of reward value.

Much of the existing HR technology literature treats AI adoption as producing outcomes along a single beneficial-to-harmful dimension, where the question is how much benefit and how much harm, rather than which pathway is activated and why (Raisch and Krakowski, 2021; Jarrahi et al., 2021). The dual-pathway model challenges this framing. The argument is not that predictive HR algorithms are good or bad but that the same underlying technology, operating through different design antecedents and organizational contexts, produces motivationally opposite outcomes. This reframing has practical consequences: it shifts the evaluative question from “should we adopt AI-powered HR systems?” to “what design and governance conditions activate the empowerment pathway rather than the hijacking pathway?” The latter question is both more tractable and more useful.

The AI ethics literature has focused primarily on fairness, bias, and transparency as the core concerns of ethical algorithmic design (Burrell, 2016; Susser et al., 2019; Zuboff, 2019). These concerns are legitimate and important. What the Neural Hijacking framework adds is a motivational dimension: ethical algorithmic design in HR contexts is not only about ensuring that the system treats employees equitably and transparently but also about ensuring that its design does not systematically erode the intrinsic motivation that makes employees capable of the adaptive, creative, self-directed work that organizations need. Fairness and transparency are necessary conditions for ethical HR AI. They are not sufficient. A system can be fair, transparent and explainable while still being optimized for engagement metrics in ways that produce motivational dependency over time. The framework makes this distinction theoretically explicit.

The algorithmic management literature has documented constrained autonomy and reduced worker agency as consistent outcomes of algorithmic direction (Kellogg et al., 2020; Lee et al., 2015; Möhlmann and Zalmanson, 2017). What it has not fully addressed is why these effects vary across employees and organizational contexts. The moderating conditions developed in Section 4.4, specifically algorithmic literacy, organizational culture and individual motivational orientation, offer three theoretically grounded explanations for this variance. They also generate testable predictions (Proposition Set 4) about the conditions under which neural hijacking effects are amplified or attenuated, giving algorithmic management researchers a more differentiated theoretical toolkit.

The most direct practical implication of the framework is for the people who build these systems. Platform vendors who optimize their products for engagement metrics, daily active users, completion rates and streak maintenance are structurally biasing their systems toward the neural hijacking pathway, often without intending to and frequently without the client organization noticing until the motivational damage is visible in turnover, disengagement or innovation decline (Kellogg et al., 2020; Shin, 2020). The framework suggests that vendors need a second optimization target running alongside engagement: a developmental outcome measure, such as skill transfer scores, employee-reported competence growth or self-initiated learning activity, that tracks whether engagement is producing genuine development or merely platform dependency.

Design features that warrant particular scrutiny are variable reward schedules, social comparison leaderboards and streak-based loss aversion mechanics. Mekler et al. (2017) found that these specific elements reliably undermine intrinsic motivation when used without commensurate increases in challenge and developmental content. Removing them entirely is not necessarily the answer: Koivisto and Hamari (2019) show that gamification elements can support motivation when they provide genuine progress information aligned with user goals. The design question is not whether to use these features but whether they are delivering information that employees actually find developmentally meaningful or simply generating the neurological conditions for continued platform use (Fogg, 2003; Eyal, 2014).

HR managers and L&D practitioners occupy a critical position in the framework: they are the organizational actors most likely to detect pathway drift and most able to respond to it before its effects compound. Three practical actions follow from the framework.

First, calibration audits that compare engagement metrics against developmental outcome data should be conducted at regular intervals and certainly after every significant platform update. If completion rates are rising while skill transfer scores are flat or declining, the system has likely drifted toward the hijacking pathway (Burrell, 2016; Raisch and Krakowski, 2021).

Second, co-design processes that involve employees in shaping their algorithmic learning environments directly support the autonomy antecedent of the empowerment pathway. Employees who have meaningful input into which goals their system optimizes for and which types of recommendations it generates are less susceptible to the choice-narrowing mechanism that P3c identifies as a driver of autonomy thwarting (Möhlmann and Zalmanson, 2017; Thaler and Sunstein, 2008).

Third, building employee algorithmic literacy through structured education about how HR systems work, what data they use, and how recommendations are generated equips employees to engage critically with algorithmic outputs rather than deferring to them (Eslami et al., 2015). This is the practical intervention that addresses the P4a moderator: literacy attenuates hijacking effects. It also reduces the opacity risk that Burrell (2016) identifies as structurally enabling algorithmic influence to operate without meaningful employee agency.

The continuum perspective in Section 4.6 carries a specific message for organizational leadership: the pathway an HR system activates is not fixed at deployment. It drifts in response to organizational pressures, and the most common pressure that pushes systems toward hijacking is executive demand for engagement metrics as evidence of HR system value. Leaders who hold their HR platforms accountable exclusively through engagement dashboards are, often unknowingly, creating the incentive for platform vendors and internal HR teams to optimize for those metrics at the expense of the developmental outcomes the system was procured to deliver (Zuboff, 2019; Kellogg et al., 2020). Redefining what counts as evidence of HR system value, to include longitudinal skill development, employee-reported autonomy and self-initiated learning activity alongside completion rates, is a governance decision that sits at the leadership level and shapes everything downstream.

The policy implications of the neural hijacking framework operate at two levels: organizational governance and regulatory design.

At the organizational governance level, the framework supports the case for mandatory psychological sustainability audits of AI systems that influence employee learning and performance. These audits would assess not only whether a system is fair and unbiased in its recommendations but also whether its design features are associated with intrinsic motivation support or erosion over time. The distinction matters because a system can pass conventional fairness audits while still systematically eroding the motivational conditions that make fair treatment developmentally meaningful (Susser et al., 2019; Yeung, 2017).

At the regulatory level, the framework aligns with and extends existing arguments for ethics-by-design requirements in algorithmic systems. Sunstein (2016) argues that the ethics of nudging cannot be evaluated solely by whether freedom of choice is formally preserved: the relevant ethical question is whose interests the choice architecture serves. Applied to HR systems, this suggests that regulatory standards for AI in workplace contexts should include requirements for developmental outcome disclosure alongside engagement metric reporting, giving employees, unions and regulators independent evidence of whether a system is empowering or exploiting its users. The right to algorithmic autonomy, including the ability to inspect, contest and opt out of algorithmic recommendations that employees experience as manipulative, is a natural extension of existing data protection frameworks and should be codified in employment law governing AI-mediated work environments.

No conceptual framework is without boundaries, and being explicit about them strengthens rather than weakens the contribution.

The framework is developed primarily in the context of formal organizational HR systems in employed work settings. Its applicability to gig economy contexts, where algorithmic management operates without an organizational HR function, is theoretically plausible given Möhlmann and Zalmanson's (2017) findings on platform workers, but requires separate empirical validation. The motivational dynamics of algorithmically managed independent workers, who lack the organizational culture moderator identified in Section 4.4, may differ in important ways.

The neuroscientific grounding of the Neural Hijacking construct rests on laboratory findings about dopaminergic reward processing (Schultz et al., 1997; Volkow et al., 2011; Robinson and Berridge, 2008) that have not yet been directly tested in HR platform contexts. The neurological mechanism is theoretically sound and supported by analogous findings in social media and gaming research (Montag et al., 2019; Alter, 2018), but direct neuroimaging evidence from workplace algorithmic environments does not yet exist. Future research using experience sampling, Electroencephalography (EEG) or longitudinal biomarker data could close this gap.

The framework also treats algorithmic systems as the primary agent of motivational influence, which risks understating the role of organizational culture, managerial relationships and peer dynamics in shaping how employees experience and respond to algorithmic management. The moderating conditions in Section 4.4 acknowledge this partial correction, but a fuller account of the interaction between human and algorithmic management influences remains for future development.

Finally, the construct of Neural Hijacking, while theoretically distinctive from existing concepts such as the undermining effect (Deci et al., 1999) and motivation crowding (Frey and Jegen, 2001), requires empirical validation to establish that it captures something measurably different from what these prior constructs already explain. The propositions in Section 5 are designed to generate that validation, but until the empirical work is done, the distinctiveness of Neural Hijacking as a construct remains a theoretical claim rather than an established finding.

This paper developed the Neural Hijacking versus Empowerment Framework to address a specific gap in the HR technology literature: the absence of a theorized motivational mechanism explaining why identical algorithmic systems produce divergent outcomes across different implementation contexts. The gap matters practically. As organizations deploy predictive HR algorithms at scale, the question of whether those systems build or deplete workforce capacity is no longer theoretical. It is a governance question with measurable long-term consequences for innovation, adaptability and sustainable performance.

The framework’s core argument is that predictive HR algorithms are not motivationally neutral tools. They are behavioral environments whose design antecedents, psychological mechanisms and organizational contexts determine whether employees move toward sustained intrinsic motivation and genuine skill development or toward surface compliance, behavioral dependency and progressive motivational erosion. This determination is not made once at deployment. It is made continuously through optimization updates, organizational metric choices and the accumulated design decisions of vendors who are frequently incentivized to maximize engagement rather than development (Kellogg et al., 2020; Zuboff, 2019).

By integrating SDT (Deci and Ryan, 1985; Ryan and Deci, 2000), Nudge Theory (Thaler and Sunstein, 2008; Sunstein, 2016) and neuroscientific research on dopaminergic reward processing (Schultz et al., 1997; Volkow et al., 2011; Robinson and Berridge, 2008), the framework offers three things that prior work in this space has not provided together. First, a theorized mechanism connecting specific design features to specific psychological processes. Second, a moderator structure that explains variance in outcomes across employees and organizational contexts. Third, a set of fifteen propositions specific enough to operationalize, non-tautological and grounded in prior empirical findings, that translate the framework into a testable research agenda.

The practical stakes are clear enough. Organizations that hold HR systems accountable through engagement metrics alone are selecting for the neural hijacking pathway without knowing it. The design decisions that maximize completion rates and daily active users are frequently the same decisions that undermine intrinsic motivation, narrow employee agency and build the kind of shallow, cue-dependent engagement that looks like productivity on a dashboard while quietly depleting the workforce’s capacity for creative, adaptive, self-directed work (Raisch and Krakowski, 2021; Deci et al., 1999). Recognizing this dynamic is the first step. Designing governance structures, audit processes and regulatory frameworks that make the empowerment pathway the default rather than the exception is the harder, more important work that follows.

The propositions developed in Section 5 constitute the immediate empirical agenda. Four broader research directions extend beyond those propositions and warrant independent attention.

The most pressing empirical need is longitudinal research tracking employees across algorithmic system updates to measure motivational change over time. Cross-sectional studies can establish associations between design features and motivational states at a single point, but they cannot capture the drift dynamic that Proposition Set 5 describes: the process by which systems shift between empowerment and hijacking as organizational pressures and optimization targets evolve (Burrell, 2016; Kellogg et al., 2020). Longitudinal designs using experience sampling methods, combined with platform behavioral data and periodic motivational assessments, would allow researchers to identify the conditions and timescales under which pathway drift occurs and to test P5a’s prediction that drift becomes detectable within two optimization update cycles.

The neuroscientific grounding of Neural Hijacking rests on established laboratory findings about dopaminergic reward processing (Schultz et al., 1997; Volkow et al., 2011) that have not been tested directly in workplace algorithmic contexts. EEG and Functional Magnetic Resonance Imaging (fMRI) protocols comparing neural activation patterns under empowerment-oriented and hijacking-oriented feedback designs would provide direct evidence for or against the reward recalibration mechanism the framework proposes. Psychophysiological measures, including cortisol response as a marker of stress and autonomic arousal as a marker of engagement quality, offer more accessible alternatives to full neuroimaging that could be integrated into field study designs with realistic sample sizes. This research direction addresses the most theoretically ambitious claim of the framework and is the most important for establishing Neural Hijacking as a distinct empirical construct rather than a relabeling of existing undermining effect findings (Deci et al., 1999; Frey and Jegen, 2001).

The moderating conditions developed in Section 4.4 are theorized at a general level. How they operate across cultural contexts is an open empirical question. Hofstede-derived research on power distance and uncertainty avoidance suggests that employees in high power distance cultures may be less likely to contest algorithmic recommendations, making the algorithmic literacy moderator less effective in those contexts (Möhlmann and Zalmanson, 2017). Employees in collectivist cultures may respond differently to social comparison leaderboard features, given that leaderboards activate both competitive and relational dynamics whose balance varies cross-culturally (Koivisto and Hamari, 2019). Cross-sector variation also warrants attention: the neural hijacking risk profile in a high-stakes professional context, such as healthcare or legal services, where algorithmic dependency carries direct performance consequences, differs from the risk profile in a corporate learning platform context. Sector-specific empirical work would allow the framework’s propositions to be tested and refined in contexts with meaningfully different motivational baselines and stakes.

Proposition P4a predicts that employee algorithmic literacy moderates the relationship between neural hijacking mechanism exposure and intrinsic motivation decline. If this moderation effect is confirmed empirically, it creates a direct intervention opportunity: structured algorithmic literacy programs that help employees understand how HR systems work, what behavioral data they use, and how recommendations are generated should attenuate hijacking effects and support more critical, agentive engagement with algorithmic outputs (Eslami et al., 2015; Burrell, 2016). Intervention research testing the design, delivery and effectiveness of such programs would connect the theoretical framework directly to organizational practice. It would also generate evidence relevant to regulatory debates about whether disclosure requirements for algorithmic HR systems are sufficient to protect employee agency, or whether active literacy education is necessary to make disclosure meaningful in practice (Sunstein, 2016; Susser et al., 2019).

Proposition P5c identifies competitive pressure on vendors to demonstrate engagement metrics as a driver of algorithmic drift toward the hijacking pathway. This is a structural rather than individual-level claim, and it requires a different research methodology: longitudinal analysis of platform design changes across vendor update histories, combined with organizational adoption data and motivational outcome measures. Policy evaluation research examining whether ethics-by-design mandates, developmental outcome disclosure requirements or psychological sustainability audit standards produce measurable shifts in vendor design behavior would provide evidence for the regulatory implications developed in Section 6.3 (Yeung, 2017; Zuboff, 2019). Without this research, policy recommendations in this space remain normatively motivated but empirically unsupported.

The Neural Hijacking construct is, at this stage, a theoretically grounded claim about a risk that predictive HR systems pose to employee motivation and workforce sustainability. Whether it describes a genuinely distinct phenomenon, one that operates through mechanisms not already captured by the undermining effect, motivation crowding theory or algorithmic management research, is an empirical question. The framework's value lies in making that question precise enough to answer.

The authors would like to thank their respective institutions for providing academic support and resources that facilitated the completion of this work. The authors also appreciate the constructive comments provided during the peer-review process, which helped improve the quality of the manuscript.