Atmospheric Regulation in the Age of Generative AI: The Sovereign Hive and the Tutor-in-the-Loop (TITL) Framework for Equity in Further Education

Synthesis: Herron (2026) introduces a conceptual framework developed through practitioner-led inquiry in an Irish Youthreach setting, centering marginalized Further Education and Training (FET) learners. The paper uses a 'Mars Gravity' metaphor to describe the intensified pressure of overlapping neurodivergent, social, and psychological burdens; when these exceed available resources, learners enter 'Barren Mars Gravity' (BMG), a state of cognitive and relational collapse. In response, the 'Sovereign Hive' is proposed as a relational learning environment functioning as an atmospheric regulator, operationalized by the Tutor-in-the-Loop (TITL) framework. Through the shame-proof 'Abditory' and a Mars Zone of Proximal Development, learners are moved from BMG toward Emancipated Mars Gravity (EMG). The work reframes Generative AI equity as atmospheric AI Regulation in Education rather than mere tool access, connecting Cognitive Offloading, Metacognition, Neurodiversity, and Human-in-the-Loop in Adult Learners contexts, and argues that systemic parity for educators is the condition that makes this restorative work possible.

Key Findings

1. Generative AI is splitting education along structural lines: learners with cognitive stability, linguistic fluency, and cultural capital reap a 'productivity dividend,' while marginalized FET learners face a 'homework apocalypse' of invisible Cognitive Offloading and AI-assisted substitution, driven by the AI Productivity J Curve.
2. The paper introduces Mars Gravity (MG) and Barren Mars Gravity (BMG) to model the compounded burdens of neurodivergence, marginalization, and domestic instability; under BMG the blank AI interface can trigger cognitive shutdown rather than Creativity.
3. The Sovereign Hive is a relational learning environment that acts as an atmospheric regulator, operationalized by the Tutor-in-the-Loop (TITL) framework and the shame-proof 'Abditory,' moving learners from BMG toward Emancipated Mars Gravity (EMG) through the Mars Zone of Proximal Development (M-ZPD).
4. Effective GenAI integration for marginalized learners is a matter of relational architecture, not technical proficiency; systemic parity for educators — preparation, recognition, and working conditions — is a prerequisite, not a supplement, to the TITL's restorative work.

The Productivity Dividend vs. Cognitive Collapse

As Generative AI permeates education, a divide has emerged between those reaping a 'productivity dividend' and those facing cognitive collapse through invisible Cognitive Offloading. The paper situates this in Further Education, where marginalized learners carry overlapping burdens that an unregulated AI environment can intensify rather than relieve. Human-AI collaboration rewards those already equipped with the cognitive stability, linguistic fluency, and cultural capital to work with machine systems, while for learners under Mars Gravity the open-ended, high-ambiguity AI interface demands initiative and self-direction that are precisely the capacities compromised under pressure.

The 'homework apocalypse' — a collapse in assessment authenticity driven by invisible offloading and AI-assisted substitution — has produced a policing climate in which teachers risk shifting from educators to detectives and students develop 'flagxiety,' an anxiety induced by fear of unreliable AI detector allegations. Neurodivergent students face the additional challenge that their cognitive uniqueness can be interpreted as misconduct. For FET learners navigating compounded pressures, neither trajectory is viable without a stabilizing intervention, and the challenge is framed as environmental rather than merely technological.

Mars Gravity and the Sovereign Hive

Using the metaphor of Mars Gravity (MG) for intensified pressure, the author describes 'Barren Mars Gravity' (BMG) as a state of cognitive and relational collapse when available resources are exceeded. This is not only a conceptual but a linguistic move: language constructs the emotional and cognitive atmosphere in which learners interpret their own capacity, safety, and Learner Agency. MG is a linguistic ecology that names a world marginalized learners already inhabit, which explains why the blank AI interface can feel hostile rather than helpful.

The 'Sovereign Hive' is a relational learning environment that functions as an atmospheric regulator, and the Tutor-in-the-Loop (TITL) framework operationalizes this regulation so human tutors remain the locus of relational and cognitive care. Grounded in an ethics of care aligned with Noddings, the Hive functions as a pre-SEL intervention, restoring the conditions under which self-awareness and self-management can meaningfully develop rather than assuming them as starting points. This foregrounds Well-Being, Neurodiversity, and Equity over purely technical integration.

The Abditory and the Mars Zone of Proximal Development

The Abditory is a protected, shame-proof digital environment for AI experimentation — a concentric relational space where the learner's voice can begin to form before it is required to perform. For learners who carry histories of institutional misrecognition, the act of beginning academic work in a visible, evaluative space carries significant psychological risk. The Abditory provides the low-visibility starting point that interrupts the shame-cycle accompanying BMG and restores the capacity to take the smallest possible academic step. It is a structural response to blocked communication and systemic educational harm, where lower-attaining students otherwise construct themselves as less worthy, less visible, and less connected.

The Mars Zone of Proximal Development (M-ZPD) recalibrates Vygotsky's Scaffolding for the age of AI: the gap to be bridged is not primarily one of knowledge but one of regulation and initiation. In the M-ZPD, the 'other' is a relational other whose role is to provide the attunement and coregulation necessary to survive the thin air of MG — the 'Digital Embrace' that prioritizes the human-to-human bond as the primary stabilizer for human-to-machine interaction.

The Tutor-in-the-Loop (TITL) as Relational Equilibrium

TITL recognizes that learners under MG cannot be expected to self-initiate in the presence of an open-ended AI interface. The tutor co-frames the task, interprets the prompt, and models the first cognitive-linguistic move, witnessing without intruding and guiding without controlling. This is an entangled intra-action in which tutor, learner, and AI are co-constituting the learning event, and the learner's willingness to try becomes a form of care offered back to the tutor. The framework is codified as a Standard Operating Procedure (SOP) moving learners through M-ZPD phases — from BMG relational entry, through shared initiation and Abditory access, to gradual release and reciprocal closure — ensuring the AI is never the lead actor but a tool safely integrated once atmospheric stability is reached.

For ESOL learners, linguistic scaffolding is not an enhancement but a prerequisite for equitable access to GenAI, since they must simultaneously decode content and language and are more vulnerable to overwhelm and misattunement. Mediated guidance can let learners think in their native tongue, and the safe space of the Abditory can bridge the Digital Divide rather than assuming universal access benefits.

Toward Emancipated Mars Gravity

The goal of the Sovereign Hive is not permanent dependency on the tutor but the transition to Emancipated Mars Gravity (EMG), a state of sustainable agency in which the learner has developed the internalized rhythms and relational stability to navigate the pressures of MG autonomously. The productivity dividend of AI becomes accessible, but it is a dividend earned through relational witnessing rather than simple automation. Movement toward EMG is demonstrated through member-checked student narratives in which scaffolded support generates moments of personal responsibility, even when a learner must visibly pull themselves out of a habitual shutdown.

The argument is ultimately one of social investment. Successful GenAI integration is not a matter of technical proficiency but of relational endurance, and the productivity dividend of the generative AI era will be realized by teachers who have been trusted, trained, and given the conditions in which to bear the relational weight of the work. National responses illustrate the principle: Norway's scaffolded age-based approach and Denmark's legal human-in-the-loop requirement, with mandated teacher oversight and legislated CPD, both validate the TITL principle.

What this means for practice

• Educators. Treat relational scaffolding as a prerequisite rather than an enhancement: co-frame the task, interpret the prompt, and model the first cognitive-linguistic move before the learner is asked to initiate with the AI.
• Educators. Protect the first academic step by giving learners a low-visibility, shame-proof space (the Abditory) to experiment in before they must perform in a visible, evaluative setting.
• Administrators. Fund the conditions for atmospheric regulation — preparation, recognition, and working conditions — because TITL's relational and cognitive labor cannot be carried by an under-supported "docile professional body."
• Policymakers. Follow the scaffolded, legislated human-in-the-loop models (Norway's age-based approach, Denmark's legal oversight requirement with mandated teacher oversight and CPD) rather than letting the "cheating" debate set AI policy.
• Researchers. Pilot the TITL Standard Operating Procedure across multiple FET settings with measures that capture movement between BMG and EMG directly, and explore technology that maps transparent Large Language Models (LLMs) use during assignment construction to reduce flagxiety.

Limitations

• The framework is propositional rather than empirically tested, developed through a single practitioner's member-checked reflective journals in one Irish Youthreach setting.
• No formal institutional ethics committee or IRB was in place at the time of the study; oversight rested on management approval, informed consent, self-chosen pseudonyms, and retrospective member checking.
• Several supporting arguments rest on unresolved evidence: the legal cases discussed (including Kato v. Palo Alto Unified School District) were ongoing at the time of writing, and the Norway and Denmark policy positions were still being implemented and evaluated.
• The high relational intensity of the model may be difficult to sustain in larger classrooms without additional staffing or revised scheduling, and transferability to other FET contexts, jurisdictions, and educator populations remains untested.

Citation

Herron, M. B. (2026). Atmospheric regulation in the age of generative AI: The sovereign hive and the tutor-in-the-loop (TITL) framework for equity in further education. EdArXiv preprint.