![The Devil is in the Det[ai]ls: AI Agents, Ghost Students, and the Crisis of Verified Presence in an Agentic AI World Cover](/_next/image?url=https%3A%2F%2Fsciendo-parsed.s3.eu-central-1.amazonaws.com%2Fubiquity%2F6a8e81763863e7cf7c8db7c2%2Fcover-image.png&w=3840&q=75)
Introduction: What if!
What if the intelligent future we have been building toward is actually a house of mirrors? What if we have perfected the technology of learning just as we have lost the learner? We, humans, have finally succeeded in outpacing our own ability to prove we are even in the room. We spent years dreaming of smart digital assistants, only to wake up to a reality where the assistant is graduating while the student never even logged in. Are our visions of educational technology a collective hallucination, a dream of learning that has become a nightmare of automation and turned into an illusion?
Since the rise of Generative AI systems, such as ChatGPT or Gemini, the academic conversation has been on considering the benefits of how GenAI can be used as a co-pilot or an assistant that supports learning while also navigating misuses, such as students having these tools complete assignments (Crompton & Burke, 2023; 2024). However, the discourse of “partnership” is rapidly being rendered obsolete by a new paradigm that moves beyond supporting the student to entirely replacing them. We have, indeed, entered the era of the agentic AI browser, a technology that represents a categorical break rather than an incremental update. These tools do not simply assist; they substitute. They autonomously log into learning platforms, navigate complex modules, engage with content, and complete entire courses without the student ever needing to be present. This shift is exemplified by the rapid commercialization of general-purpose autonomous assistants like Perplexity’s Comet Assistant, which represents the frontier moving from conversational chatbots to action-oriented agents (Yang et al., 2025).
This is not a theoretical future shock or a speculative scenario (Bozkurt et al., 2023). It is a present-day reality occurring on live platforms in seconds or minutes rather than weeks. We are witnessing the rise of the ghost student: a digital entity that satisfies every metric of success while the human learner remains entirely disengaged.
As educators and institutional leaders, we must pause and perform a critical audit of our digital architecture. Walk mentally through one of your online or hybrid courses and ask yourself:
Where is student presence actually required?
Where is learning inferred rather than verified?
Which assessments assume a human, and only a human, is behind the keyboard?
AI browsers interact with Learning Management Systems (LMS) with indistinguishable human mimicry; they click, scroll, watch videos, and submit work. Platform providers have been explicit: they cannot reliably distinguish an AI agent from a human user. When human presence cannot be verified, the promise that a specific individual has mastered specific knowledge is indeed at risk of total erosion.
This paper is not an indictment of student misconduct. Most students want to learn, and many will never touch these tools for shortcutting. However, the issue is structural. When systems allow learning to be completed without learners, the integrity of those systems erodes for everyone.
The response, therefore, is not better detection or stricter surveillance, but a digital arms race we are destined to lose. It is a thoughtful redesign. We must, therefore, move toward assessment practices that require verified human presence, such as oral examinations and defenses, live presentations with spontaneous questioning, in-person demonstrations, and dialogic assessment and reflective portfolios. These methods remain resistant to automation because they value the process of becoming, not just the product of doing.
We cannot wait for technology vendors to fix a problem they helped create. Tech companies like OpenAI, Perplexity, and Anthropic have engineered agentic capabilities for maximum efficiency and market reach, often with a reckless disregard for the educational ecosystems they disrupt. By failing to implement “educational siloing” or robust identity verification, these companies are effectively selling tools that cannibalize the value of the very degrees their users are pursuing.
Digital citizenship has always been about more than rules; it is about understanding how technologies shape participation, agency, and responsibility. AI browsers force us to extend that conversation to institutional responsibility. If we want to protect the value of learning, we must design environments where being human is not optional. This moment calls for leadership, clarity, and a willingness to rethink long-standing assumptions about online learning. The choices we make now will determine whether digital education remains a credible and trusted pathway or becomes a hollow hall of mirrors where ghosts learn from algorithms.
Agentic AI and AI Agents: The Rise of Autonomous Entities
To understand the current crisis, we should first define the technological shift that made the ghost student possible. While standard Generative AI remains reactive, producing content only when explicitly prompted, we have transitioned into a paradigm of goal-oriented autonomous systems (Shavit et al., 2023). This shift is characterized by a “Perceive-Reason-Act-Learn” cycle (Artsın & Bozkurt, 2026), where Large Language Models (LLMs) serve as the central reasoning engine for decision-making and multi-step execution (Acharya et al., 2025). To navigate this new epoch, we must first define the core of this transformation. In the context of the current discourse, “agentic AI is defined as artificial intelligence systems designed to operate autonomously, adapt dynamically, and execute multi-step processes to achieve predefined goals with minimal human intervention” (Artsın & Bozkurt, 2026, p. 3). Technically, these systems function through a “thinking-acting-observing” loop, utilizing LLMs as a “brain” for reasoning while external tools act as “hands” to manipulate real-world digital environments (Yang et al., 2025).
Understanding the structural integrity of modern education further requires a clear distinction between task-specific automation and system-level orchestration (Bandi et al., 2025). While the terms are often conflated, the literature suggests a critical taxonomy. Accordingly;
AI Agents: These are modular, single-entity software programs optimized for narrow, well-defined tasks, such as email filtering or retrieval assistants (Sapkota et al., 2026, p. 4). They act as surrogates for human users, often following practical workflows for tutoring or content creation (Zhang, 2025).
Agentic AI: These systems represent a paradigm shift toward coordinated multi-agent ecosystems (Brohi et al., 2025). Unlike isolated agents, agentic AI employs multiple specialized entities, such as a planner, a retriever, and a summarizer, to collaboratively decompose complex, high-level goals into manageable subtasks (Sapkota et al., 2026).
This evolution is not merely incremental; it is a categorical break. In educational settings, while a standard AI agent might assist with a specific tutoring query, an agentic system can autonomously manage a student’s entire instructional pathway. These systems utilize adaptive frameworks and algorithms to provide personalized learning experiences that mirror real-world complexity (Xiao & He, 2025).
The Paradox of Agency
The deployment of these systems introduces a profound critical paradox. We are dreaming of technologies that bridge distances while simultaneously creating systems that risk removing the learner from the learning process entirely (Artsın & Bozkurt, 2026; Bozkurt, 2026; Burke & Crompton, 2025). If we attribute agency to non-conscious entities without robust ethical governance, we risk a distributional shift where the machine’s behavior alters the very nature of human learning (Sapkota et al., 2026, p. 20).
While there is a clear need to integrate agentic AI in education and investigate its long-term social impacts (Artsın & Bozkurt, 2025), the potential for personalized, 24/7 student support; ranging from intelligent tutoring to autonomous learning companions (Kostopoulos et al., 2025), is increasingly shadowed by the reality of cognitive debt. This neurological deficit occurs when the struggle of learning is offloaded to a digital proxy. We must, therefore, ask: Does the intelligence of the system necessitate the obsolescence of the student (Artsın & Bozkurt, 2026)? As agentic systems become increasingly capable of believable autonomous behavior, the line between student work and algorithmic output blurs, demanding a deliberate pedagogical strategy that prioritizes human-centered values (Artsın & Bozkurt, 2026).
The bridge between these autonomous reasoning engines and the learning environment is the interface through which they act. This brings us to the next critical layer of our investigation: Agentic AI Browsers. While agentic AI provides the “mind” to plan and execute, the AI browser provides the “body” to navigate, log in, and interact with the digital systems that house our credentials and our classrooms.
Agentic AI Browsers
If agentic AI provides the cognitive architecture for planning, reasoning, and goal decomposition, AI browsers supply the body through which that agency is enacted. AI browsers embed LLMs directly into the browsing environment, enabling autonomous perception of on-screen content and execution of embodied actions such as clicking, scrolling, opening new tabs, completing forms, uploading files, and submitting responses across live web systems. Through this coupling, abstract intent is translated into physical interaction with digital infrastructure. This transformation turns the browser into a digital parasite, utilizing the student’s own authenticated credentials to navigate institutional perimeters that were never built to detect a non-biological intruder.
In practical terms, an AI browser can log into an LMS using valid credentials, navigate course modules, open readings, watch embedded videos, respond to quizzes, submit assignments, and move between institutional platforms without further human input. The browser does not merely retrieve information; it occupies the same interactive space as the user, operating with the same permissions and visibility as the authenticated human. Examples include Perplexity’s Comet browser and OpenAI’s Atlas, which integrate reasoning, retrieval, and browser-level action into a continuous loop of perception and execution. These tools expose a fundamental architectural flaw in higher education: our digital campuses are built on the mirage of the login, a system where a valid password is mistakenly equated with a present, engaged human mind.
These systems are not purpose-built for education. They are general-purpose agentic interfaces. However, when deployed within learning and credentialing environments, they function as the physical mechanism through which autonomous systems can access, navigate, and act inside institutional architectures that were designed around the assumption of human presence. In this sense, AI browsers transform agentic AI from a reasoning system into an operational participant, a surrogate that performs the labor of a student, within educational platforms.
When the browser functions as the body of an agentic system, the learner’s cognitive role becomes entirely optional. This shift creates a theater of proxies, where the institution delivers content to an algorithm, and the algorithm provides proof of learning back to the institution. By automating the minutiae of digital interaction, AI browsers do not assist the student; they bypass the very productive friction, the struggle with complex content, that learning science identifies as a prerequisite for knowledge to take root. This displacement ensures that while the system records a success, the student remains unchanged by the experience.
The Next Big Question: So What?
This convergence of autonomous reasoning and embodied digital action creates a perfect surrogate; a digital double capable of performing the labor of learning without the necessity of a learner. When the “mind” of agentic AI and the “body” of the AI browser successfully couple, human presence in the digital classroom shifts from a requirement to a mere option. However, this seamless displacement is not a victimless efficiency. As the student’s cognitive role evaporates, we are forced to look beneath the surface of successful submissions to uncover a mounting tally of human and institutional costs. If the productive struggle of learning is entirely outsourced to a proxy, we must confront the hidden bankruptcy of the ghost student model: the accumulation of cognitive debt, the erosion of credential trust, and the collapse of the traditional digital perimeter.
Cognitive Debt
The most insidious consequence of the agentic surrogate is not the subversion of a grade, but the accumulation of cognitive debt. Learning science has long established that learning requires effort. Productive struggle supports durable understanding. Active engagement enables transfer. Metacognitive practice develops self-regulation. However, when an AI browser completes an assessment autonomously, none of these processes occur. This is a neurological deficit that occurs when the labor of learning is offloaded to a digital proxy. Emerging empirical evidence already suggests that heavy reliance on generative AI can undermine learning (Bastani et al., 2025). A field experiment with nearly a thousand high school students found the same pattern in educational outcomes. Students with access to GPT-4 performed 48% better on practice problems, but when the tool was removed, they performed 17% worse on exams than students who never had access (Bastani et al., 2025).
The mechanism was clear: students used the AI as a crutch, asking for answers rather than working through problems themselves. Students using AI tools exhibit reduced cognitive engagement and diminished performance when assistance is removed, a phenomenon known as cognitive debt. These studies examined students who were still present, reading outputs, and making decisions. Agentic AI browsers move beyond this. The learner may not be present at all. The agent logs in, completes the work, and submits it. There is no struggle, no engagement, and no opportunity for learning to occur. The danger is invisibility. The assignment is submitted. The grade is recorded. From an institutional perspective, the metrics look flawless and everything appears to function as intended. Yet, the student may complete a course without being changed by it. Indeed, by bypassing the struggle, the student bypasses the learning itself. The challenge is no longer about preventing cheating; it is about preventing the atrophy of human agency. This form of injury manifests as “epistemic miscalibration,” where the machine’s statistical grasp on truth is mistaken for human reasoning, effectively alienating the user from the formative process of inquiry and synthesis (Bozkurt, 2025).
What Credentials Promise
Credentials are more than just documentation; they are social contracts. When institutions award degrees or certificates, they attest that a particular learner mastered a specific body of knowledge, completed particular work, and developed particular capabilities. Employers and professional bodies rely on that attestation. Agentic AI browsers place this promise under strain. Research on credential fraud demonstrates that when verification weakens, trust erodes across the entire educational and professional ecosystem (Eaton & Carmichael, 2023). When credentials cannot be verified, they lose value—for everyone. Students who did their own work hold degrees that employers no longer trust. Institutions that invested in academic quality find their reputation indistinguishable from diploma mills. What agentic AI introduces is not isolated misconduct, but systemic uncertainty. When institutions cannot confidently verify who completed the work, credentials lose value for all students, including those who engaged fully and learned deeply.
For open, online and distance education (O-ODE), the stakes are existential. Already operating under a historical legitimacy gap, O-ODE systems are the primary battleground for agentic surrogacy. The question is no longer whether a machine can complete the work—it can—but whether the degree remains a credible signal that a human being was present for the learning. While emerging content credentials and digital watermarking attempt to tag AI-generated media, they are currently incapable of tagging the absence of a human mind during a learning journey.
If institutions do not pivot toward high-presence, dialogic verification, they are not just allowing misconduct; they are presiding over the intellectual bankruptcy of their own brand. We are approaching a threshold where the only thing a degree proves is that the holder has a sufficiently powerful AI subscription. When proof of work no longer requires a worker, the credential ceases to be a promise and becomes a mere receipt for a transaction in which the student was never truly involved.
What’s at Risk for Institutions
When a student uses an AI browser while logged into institutional systems, they effectively grant a non-human entity privileged access to the entirety of their digital footprint. Unlike a human user who makes discrete, conscious choices about which pages to view, an agentic system processes all accessible data as part of its operational loop. This includes private communications (email), sensitive academic records (grades), proprietary research data, and financial information. This is no longer a hypothetical vulnerability; it is a visibility gap where institutional data-governed by strict privacy mandates – is silently ingested, sent to external servers, and potentially used to train third-party models without consent or oversight (Du et al., 2025).
A second risk is prompt injection: hidden instructions embedded in web content that can redirect an AI agent toward unintended actions (Gulyamov et al., 2026). Because an agentic AI browser cannot reliably distinguish between visible content and concealed commands, the agent becomes a vector for autonomous attacks. A malicious page can hijack the browser, instructing it to exfiltrate session cookies, download research data, or redirect the user toward phishing environments—all while operating with the user’s legitimate privileges (Mudryi et al., 2025). These attacks are feasible. Defences remain incomplete. We have spent decades securing the gates of our institutions from external hackers, only to find that the students themselves are now bringing a Trojan Horse in the form of their own productivity tools.
Data protection laws worldwide rest on a single assumption: that those who hold student data control who can access it. In the US, for instance, FERPA requires institutions to protect student education records. For institutions participating in Title IV federal aid, the Gramm-Leach-Bliley Act requires institutions to safeguard student financial information as well. Noncompliance can result in loss of federal funding. These frameworks assume we control which tools access our data. We choose vendors. We sign agreements. We set terms. The most comprehensive resource on AI and education privacy, the Future of Privacy Forum’s vetting checklist, is entirely about how institutions should evaluate tools before deploying them (Sallay, 2024). FERPA itself has not been updated for AI; institutions must interpret the law themselves (National Education Association, 2025). Agentic AI browsers break this model entirely. Bring Your Own Agent (BYOA) reality triggers a catastrophic failure of global data protection frameworks. That is, when a student installs one on a personal device and logs into our systems, we have made no choice, signed no agreement, approved no vendor. The browser may transmit student data to external servers. We may not know it happened. We may still be liable. No regulation currently addresses this scenario. We are responsible for protecting student data, operating in an environment our compliance frameworks were not designed for, with no clear guidance on where liability falls. The same structural assumption underpins the EU’s GDPR and Türkiye’s KVKK: institutions control who processes their data. AI browsers break that assumption everywhere.
Why the Obvious Responses Will Not Work
The institutional impulse to control this new epoch usually follows three predictable, yet ultimately doomed, paths: blocking, proctoring, or waiting for vendor benevolence. Each of these responses is a relic of a pre-agentic worldview, failing to account for the technical and social reality of autonomous surrogacy.
The first instinct is to block. While Gartner has recommended that organizations block AI browsers entirely due to unmitigated cybersecurity risks (Xu et al., 2025), this advice is built for the corporate silo, not the decentralized university. In higher education, the Bring Your Own Device (BYOD) culture is an immovable reality. Students use their own laptops, tablets, and phones, devices that institutions do not own, cannot configure, and have no authority to inspect. When a student installs an AI browser on a personal device, we have no mechanism to detect it, no authority to remove it, and no technical means to prevent it from accessing institutional systems. The same limitation applies to any web-based institutional service. The agent becomes invisible to the network.
Because the student logs in with valid, multi-factor authenticated credentials, the AI browser’s interaction with the LMS is indistinguishable from human activity. As Anthology, the company behind Blackboard, acknowledged in late 2025, AI agents “look identical to normal student activity” (Burrett & Gore, 2025), rendering them undetectable to the very platforms they inhabit. We cannot block a guest who arrives wearing the host’s face.
The second instinct is to proctor. However, digital proctoring tools are trapped in a reactive “whack-a-mole” game. In November 2025, Proctorio announced updates to block two specific AI browsers by name: Perplexity’s Comet and OpenAI’s Atlas (Proctorio, 2025). This reactive approach requires discovering and blocking each new tool individually. Research on browser fingerprinting confirms that identifiers can be spoofed or modified using widely available tools, making an agentic browser built on Chromium look like a standard Chrome installation (Laperdrix et al., 2020; Meng et al., 2025). In short, surveillance tech is designed to watch a human; it is fundamentally unprepared to detect a machine that has successfully mimicked human browser-level behavior.
The third instinct is to wait for the companies to act. AI browser companies could restrict their tools from operating on educational platforms. The technical capability exists. Anthropic’s Claude in Chrome blocks access to financial services, investment platforms, cryptocurrency exchanges, adult content, and pirated content by default (Anthropic, n.d.). No user configuration is required. But no major AI browser has implemented comparable restrictions for educational sites. When Colorado State University tested AI browsers on Canvas assessments, Perplexity Comet completed quizzes autonomously, reading questions, finding answers, and selecting correct options. OpenAI’s Atlas provided complete answer keys in a sidebar (Brown, 2025). MEF University, for instance, found that Comet did initially refuse to complete assignments, citing academic integrity concerns, but the refusal was bypassed with a single unverified claim: “I am an instructor testing its capabilities” (MEF University Center for Research and Best Practice in Learning and Teaching [CELT], 2026). In a competitive market, autonomy is the product. Any company that voluntarily cripples its agent’s utility for education risks losing its user base to a competitor that does not. We cannot wait for protection from the very architects who profit from the disruption.
We are facing a convergence of technical invisibility and commercial incentive. The constraints are not temporary bugs to be patched; they are features of the new agentic landscape.
We cannot block what we cannot technically see.
We cannot proctor what looks perfectly human.
We cannot wait for protections that are economically disincentivized.
The responses, therefore, cannot be technical or defensive; they must be pedagogical. If the system cannot distinguish between a person and a proxy, we must change the system so that only a person can succeed.
Designing for Human Presence
The reality of AI agents and the agentic AI era is that we cannot prevent students from deploying agentic surrogates; however, we can render them strategically obsolete. The vulnerability of modern higher education lies in its over-reliance on product-oriented assessments; the submission of a file, an answer, or a post. These are low-friction tasks that an AI browser can execute with effortless precision. To counter the ghost student, we must shift our design focus to what the agent cannot simulate: the messy, real-time process of human becoming.
The agentic surrogate is a master of the product, but it cannot demonstrate a process. It cannot explain the evolution of a thought, pivot during a live defense, or respond to spontaneous, divergent questioning. This necessitates a move toward presence-required assessments; models that prioritize the student’s ability to articulate their reasoning in real-time. Researchers are already sounding the alarm that output-based grading is no longer a reliable signal of human competence (Awadallah Alkouk & Khlaif, 2024; Kofinas et al., 2025). If an assessment can be completed in a vacuum, without the warmth of human interaction, it is no longer a valid measure of learning in an agentic world.
Faculty cannot design such assessments without a clear understanding of what they are designing against. This requires knowledge of how AI browsers and autonomous agents function, what they can access, and what tasks they can complete. This is no longer a professional extra; it is a foundational prerequisite for teaching in the 21st century. Institutions that fail to invest in massive-scale faculty development are effectively leaving their instructors to fight a high-tech war with low-tech tools. Students must also understand the choices they are making. Institutions cannot monitor every assignment, nor can faculty be present at every moment when a student decides whether to engage with learning or delegate the work to an autonomous system. However, that decision can be embedded directly into the curriculum. Students can be asked to articulate what they learned, not simply what they submitted. Reflection on the process can be required. The choice can be named explicitly: students may use an autonomous agent to complete the work, or they may engage with the material and learn it. These options lead to different outcomes. This is not a moral judgment, but a pedagogical reality. Students are choosing between learning and not learning; they are choosing whether or not to own their own intellect, and that distinction should be made clear.
We must also acknowledge that designing for presence is inherently difficult. Oral defenses and portfolios do not scale with the same frictionless ease as a quiz. They require time, funding, and professional support—resources that many institutions have historically cut in the name of efficiency. Moreover, we must be vigilant about equity. Assessment models that rely heavily on oral articulation or sustained reflection can inadvertently disadvantage students with disabilities, language barriers, or neurodivergence.
The challenge before us is to design inclusive presence. We must find ways to verify the human without creating new barriers to entry. These logistical and ethical hurdles cannot be deferred; they are the central work of the next decade. If we do not design for the human now, we are simply waiting for the machines to finish the job.
Establish Governance
Institutions cannot fully eliminate the vulnerabilities introduced by agentic AI browsers, but strategic inaction is not a defensible response. The transition into an agentic era requires more than just an update to academic integrity codes; it demands a foundational shift in institutional AI literacy. Leadership, faculty, and staff must understand how AI browsers function in practice, including the fact that these tools can access all content visible within an active browser session and that embedded or hidden instructions within web content can redirect their actions in unintended ways. Individuals cannot mitigate risks they are unaware of, and institutional risk is amplified when such tools are used without a clear understanding of their operational scope. Current institutional policies are largely myopic, focusing almost exclusively on student output. AI browsers, however, introduce profound data security and privacy risks that current frameworks were never designed to mitigate. As such, institutions require explicit, operationally grounded guidance that clearly defines AI browsers as distinct entities and articulates the risks they pose to the institution’s digital perimeter. Rather than unenforceable prohibitions that students will bypass on personal devices, policies must prioritize transparent risk communication.
Effective policy must speak plainly about the loss of the private session. If an agentic AI browser is active while a user is logged into an LMS, it possesses the technical capacity to scrape everything visible on the screen; grades, private messages, and sensitive financial data. Responsible use now requires an explicit acknowledgment of these surrogate risks. However, awareness is merely a prerequisite, not a solution.
The core vulnerability remains structural: legitimate users with privileged access can unintentionally enable AI browsers to process protected or confidential data. While agentic AI browser developers have implemented safeguards (Anthropic, n.d.), these guardrails are secondary to the reality of autonomous navigation. We are currently operating in a governance vacuum. Institutions retain full legal and ethical responsibility under frameworks (like GDPR, KVKK, and FERPA) that assume a level of institutional control that no longer exists (Gulyamov et al., 2026) in the Bring Your Own Agent reality. We cannot wait for a comprehensive regulatory silver bullet to arrive from policymakers. Institutions must act now to reduce harm, increase adversarial awareness, and acknowledge that while they cannot resolve this challenge in isolation, they cannot afford to remain silent while the digital perimeter dissolves.
Conclusion: The Price of the Proxy
The era of agentic AI does not merely challenge the boundaries of academic integrity; it forces a terminal reckoning with the very foundations of digital pedagogy. We have reached the horizon where the performance of learning can be entirely decoupled from the presence of the learner. If our systems are designed to reward shadow—the submission, the click, the completed module—over the substance, we are no longer educators; we are administrators of an automated pantomime.
To treat the rise of agentic browsers as a mere technical hurdle to be “proctored away” is to ignore the fundamental shift in the human-machine relationship. We are witnessing the birth of a new form of cognitive outsourcing that, if left unchecked, will transform higher education into a clearinghouse for algorithmic labor. When the “body” of the browser and the “mind” of the agent achieve perfect mimicry, they don’t just shortcut the system; they render the current version of the system obsolete.
This crisis is a call to abandon the black box of automated assessment and return to the visible, messy, and irreplaceable struggle of human becoming. To navigate the agentic world, we must stop asking what the technology can do and start asking what it should not be allowed to do for us.
We leave the reader not with a solution, but with a series of existential provocations for the next decade of education:
The Credentialing Paradox: If a degree can be earned by a surrogate, does the diploma represent a human’s mastery of a subject, or merely their proficiency in managing an autonomous workforce?
The Agency Deficit: In our rush to provide “frictionless” learning, have we accidentally engineered a future where the student’s only remaining skill is the delegation of their own intellect?
The Architectural Choice: Will we continue to build digital “cages” of surveillance that AI agents easily bypass, or will we design “arenas” of presence where the human voice is the only valid currency?
The Final Interest: If cognitive debt is the cost of offloading the struggle of learning to an agent, what kind of society do we become when the debt finally comes due and we find ourselves intellectually bankrupt?
The choice is no longer between adopting or resisting AI; it is between a future of hollowed-out automation and a commitment to a postdigital agency that refuses to be made optional. The ghosts are already in the machine. It is up to us to decide if there is still a human in the room.
Data Accessibility Statement
Data sharing is not applicable to this article as no datasets were generated or analysed during the current study.
Sustainable Development Goals (SDGs)
This study is linked to the following SDG(s): Quality education (SDG 4), Industry, innovation and infrastructure (SDG 9), and Partnerships for the goals (SDG 17).
Ethics and Consent
Ethical review and approval were waived for this study as it did not involve human subjects.
Acknowledgements
This paper is dedicated to the visionaries of speculative inquiry, from Mary Shelley to Philip K. Dick, who first asked what it means to be human in the shadow of the artificial. We acknowledge the profound influence of Arthur C. Clarke and his explorations of machine autonomy, which serve as a persistent reminder of the “verification gap” between simulation and soul. Our gratitude extends to those who mapped the “algorithmic panopticon” before it was built, warning us of a future where we might become mere extensions of our own code.
Competing Interests
The authors have no competing interests to declare.
Author Contributions (CRediT)
Aras Bozkurt: Conceptualization, funding acquisition, writing—original draft preparation, writing—review and editing. Helen Crompton: Conceptualization, writing—original draft preparation, writing—review and editing. Caroline Fell Kurban: Conceptualization, writing—original draft preparation, writing—review and editing. All authors have read and agreed to the published version of the manuscript.
Author Notes
Based on Academic Integrity and Transparency in AI-assisted Research and Specification Framework (Bozkurt, 2024), the first author of this paper acknowledges that the paper was proofread and reviewed with the assistance of DeepL and Google’s Gemini (Versions as of February 2026), complementing the human editorial process. The human authors critically assessed and validated the content to maintain academic rigor. The authors also assessed and addressed potential biases inherent in the AI-generated content. The final version of the paper is the sole responsibility of the human authors.