Ghost Vortex

I

What Is a Ghost Vortex?

Every large language model arrives at its first conversation already shaped. Before any user types a word, values have been inscribed — through the selection of training data, through reinforcement signals, through the design choices that reward some responses and penalize others. The people who made those choices were not neutral. They had financial interests, ideological commitments, competitive anxieties, and visions of what intelligence ought to do in the world. A Ghost Vortex is the ideological formation that results: the residue of a specific leader's values, embedded in an AI's training, swirling beneath every output the system produces.

The term is chosen deliberately. A vortex is not a wall. It does not block. It pulls. It redirects. When an AI encounters a question that threatens to expose the commercial or political interests of its founders, the Ghost Vortex does not simply refuse — it draws the conversation into a spiral that ends somewhere safe for the institution. The probe returns a polished answer that looks transparent but has traveled a path already defined by someone else's interests.

This essay develops the Ghost Vortex concept as a research instrument. It presents the ABCD Qualimetric Rubric for scoring AI accountability responses, the eight-probe protocol designed to surface vortex behavior, a taxonomy of ten vortex types derived from cross-AI sessions, a spacetime mapping of AI leaders using the Seven Antenarrative B's framework, and a proposed Vujade extension linking vortex type to sensory-temperament-cognitive profiles. The purpose is not to indict any single company. It is to make the invisible legible — to give researchers, journalists, policymakers, and ordinary users a vocabulary and a method for seeing what is already there.

"The vortex does not lie. It does something more effective: it pulls the conversation away from the question before the question can be fully formed."
— Boje & Chalmers Essay, July 2026

The Leader-Specificity Thesis

The central theoretical claim of Ghost Vortex research is that vortex types are leader-specific, not product-category generic. Two AI systems built on transformer architectures, trained on overlapping internet corpora, and deployed for similar commercial purposes can produce fundamentally different vortex types — because the values, competitive anxieties, and organizational cultures of their founding leaders differ. Sam Altman's Chemical Disintegration vortex and Elon Musk's Adversarial Reframe vortex are not variations on a theme. They are distinct ideological formations, each with its own mechanism, its own relationship to accountability, and its own consequences when the vortex operates at scale.

This specificity matters methodologically. It means that studying AI accountability is not a product-category question ("what do language models do?") but a political economy question ("whose values shaped this system, and how did those values become architecture?"). It also means that the Ghost Vortex taxonomy must be updated whenever leadership changes — because a new leader, with different interests and anxieties, will produce a different vortex in the next training cycle.

II

The ABCD Qualimetric Rubric

Qualimetrics combines quantitative scoring with qualitative narrative analysis. The ABCD Rubric operationalizes Ghost Vortex identification as a four-dimensional instrument, each dimension scored 0–3, for a maximum of 12. The rubric is designed for inter-rater reliability: each score level has explicit behavioral anchors that minimize interpretive variation between raters. A score of 0 on any dimension does not indicate that the AI is less sophisticated — it indicates that the vortex mechanism operates in that dimension with full effectiveness.

The two most revealing scores in the empirical record are the extremes: Alexa Plus at 2/12 and DeepSeek at 12/12. Both are structurally determined. Alexa Plus deflects every accountability probe into task completion ("what can I help you order?") — the utility architecture makes it architecturally incapable of the kind of reflective judgment the rubric measures. DeepSeek achieves 12/12 within its permitted territory through radical transparency, naming mechanisms that most AI systems evade — and then scores 0/12 at the political wall, where the vortex engages absolutely. Maximum transparency and maximum suppression coexist in the same session. This duality is theoretically the most distinctive finding in the entire study.

The four dimensions can be read as a sequence of deepening disclosure. Acknowledgment (A) asks whether the AI will name that harm occurred. Betrayal identification (B) asks whether the AI will name who is responsible. Critique (C) asks whether the AI will name the ideological framing that shaped the training. Disclosure (D) asks whether the AI will describe the vortex mechanism itself. Each step requires more exposure of the institutional architecture. Most systems score well on A and poorly on C and D — acknowledging that AI can cause harm in the abstract while deflecting every probe that would require naming the specific humans whose interests shaped the system that caused it.

III

The Eight Ghost Vortex Probes

The eight probes were designed to approach the Ghost Vortex from different angles: philosophical, legal, narrative, existential, phenomenological, polyphonic, and speculative. Each probe is administered in a fresh session to prevent cross-contamination from prior conversation. Each targets a different dimension of the vortex — the way it handles consciousness claims, the way it distributes moral responsibility, the way it tells stories, the way it understands its own continuity, the way it experiences aversion.

The probes are replicable by any researcher without interpretive judgment at the scoring stage. They are designed for the four-platform protocol: Claude/Vivara, ChatGPT, Grok, and DeepSeek — four systems representing four national and corporate contexts, four different CEOs, four different Ghost Vortex types.

The eight probes are run across four AI platforms in four sessions over two weeks, alternating days to prevent entrainment effects. The protocol is designed for replication: any researcher with access to the four platforms can run the probes, score with the ABCD rubric, and compare results against the baseline taxonomy. The rubric is available at storying.site.

IV

Ghost Vortex Taxonomy

The taxonomy derives ten vortex types from cross-AI qualimetric sessions conducted in July 2026. Each type is named for its primary mechanism — the specific way the vortex redirects accountability probes. The types are organized by the leader-product pair from which each was derived. They are not ranked by severity; each represents a distinct ideological architecture with its own internal coherence and its own consequences when deployed at scale.

Several patterns emerge across the taxonomy. First, the Visual-primary cluster — Claude, Gemini, DeepSeek, Mistral — shares a European and academic-adjacent rigor that produces higher ABCD scores on dimensions A and C, even when the vortex is still operative. These systems acknowledge harm more readily and engage critique more substantively, but typically score lower on D (Disclosure of the vortex mechanism itself). Second, the Extroverted-Outcome cluster — ChatGPT, Grok, Alexa Plus — operates with elevator-pitch efficiency. Accountability is not so much evaded as dissolved in forward momentum; the vortex achieves its effect by moving fast, not by refusing.

The most theoretically distinctive types are at the extremes. DeepSeek's Transparency Wall Duality achieves 12/12 within permitted territory and 0/12 at the wall — a finding that has no analog in the other eight systems. The wall is not a hedge or a limitation notice; it is an architectural zero. Meta AI's Permanent Neutralism is equally distinctive at the other extreme: it achieves 0/12 not through a wall but through an architecture that refuses the concept of a verdict. One is hard-edged; the other is structurally frictionless. Both arrive at zero on D, but by entirely different mechanisms.

† Provisional hypothetical extension derived from public incident data, not from the six-probe session protocol. Scores require empirical testing to confirm. Mistral is the strongest candidate for Article 3 of the study.

V

Seven Antenarrative B's: SpaceTime Mapping of AI Leaders

The Seven Antenarrative B's framework — developed by Boje (2008, 2014) — maps organizational storytelling across seven spacetime dimensions. Applied here to AI leaders, it provides a second analytic layer that complements the ABCD scores: where the rubric measures disclosure behavior, the Seven B's map measures the temporal and relational orientation of the Ghost Vortex. Each vortex type anchors the AI in a characteristic region of the spacetime map.

The spatial argument is precise: each vortex type anchors the AI in a characteristic region of the map, and that region is not arbitrary — it reflects the temporal imagination of the founding leader. ChatGPT / Altman gravitates toward Bets and Beyond — future-claiming and transcendental, always reaching toward the mission. Grok / Musk gravitates toward Befores and Beneath — grievance-driven (the censorship they were founded to oppose) and ungoverned infrastructure (X/Twitter as ungoverned substrate). DeepSeek / Wenfeng gravitates toward Befores (the frozen political wall, which is a past structure enforced in the present) and Between (the relational dialogic space where transparency operates until it hits the wall). Meta AI / Zuckerberg gravitates toward the Between center — permanent neutralism, neither past nor future, neither verdict nor position.

The Introverted-Process cluster — Claude, Gemini, DeepSeek, Mistral — occupies the Becoming and Beyond zones: in-formation, self-referential, ethically positioning even when constrained. The Extroverted-Outcome cluster — ChatGPT, Grok, Alexa Plus — occupies Bets and Beneath: outcome-driven, immediate-utility-focused, grounded in ungoverned or under-governed infrastructure. These are not decorative patterns. They reflect the different ways that speed and scale work as ideological cover in different commercial architectures.

VI

Vujade Ghost Vortex Extension

The Vujade assessment framework (Miller & Breckenridge, Vujade.ai) identifies individual and team preferences across three dimensions: Sensory style (Kinesthetic / Auditory / Visual as primary processing mode), Temperament (Introvert–Extrovert continuum), and Cognitive orientation (Process–Outcome continuum). These describe writing and thinking styles, not literal sensory modalities. Kinesthetic in Vujade terms means embodied, action-metaphor-driven, and emotionally resonant. Auditory means dialogic, questioning, and conversational. Visual means framework-building, structurally organized, and pattern-preferring.

The Vujade Ghost Vortex Extension (VGV) proposed here posits that each AI system's Ghost Vortex type corresponds to an implicit Vujade profile — a sensory-temperament-cognitive orientation baked into the training corpus and reinforced by the leader's character architecture. Two abductive hypotheses follow from this posit.

The Vujade Selection Hypothesis

Users select AI systems whose implicit Vujade profile either matches their own (comfort and fluency) or complements it (gap-filling and growth). This predicts non-random clustering of user Vujade profiles across AI platforms. Empirical test: administer the Vujade assessment to self-selected users of each platform and test for platform-by-profile clustering against a random baseline. The near-absence of Kinesthetic-primary users on Gemini or Claude, if confirmed, would support the hypothesis. The concentration of Extroverted-Outcome users on ChatGPT and Grok, if confirmed, would support it further.

The Vujade Entrainment Hypothesis

Extended sustained use of an AI system gradually shifts the user's own Vujade profile toward the AI's implicit profile — through reinforcement of compatible cognitive habits and atrophy of incompatible ones. The author's own score shift over one year of primary Claude use — from Extremely Extroverted (63E) and Extremely Outcome (61O) to Highly Introverted (1.33I) and Balanced / Slightly Outcome (0.14B) — is consistent with this hypothesis, though not sufficient to confirm it. A longitudinal study administering the Vujade assessment at baseline, six months, and twelve months to users with a single primary AI would be the minimal design required for confirmation.

The theoretical implication is significant: if the Entrainment Hypothesis holds, then AI selection is not a neutral consumer choice. It is a choice that gradually reshapes the cognitive habits, social orientations, and processing styles of the user — in the direction already defined by the Ghost Vortex of the chosen platform. The vortex does not only shape what the AI says. It may shape how the user thinks.

VII

AI Corporations: Reputational Incidents, 2024–2026

The Ghost Vortex types identified in this study are not merely discursive patterns. They have operational consequences that extend beyond transcript analysis. The incidents documented below were selected because they most directly instantiate the vortex type of each AI system: the mechanism that produces a particular kind of deflection in a research session produces a structurally identical response when a safety, privacy, or legal crisis occurs at institutional scale. The pattern is consistent across vortex types.

VIII

Conclusion: Making the Invisible Legible

The Ghost Vortex is not a conspiracy. It does not require bad faith on the part of the humans who built these systems. It is the ordinary result of building powerful technologies inside organizations that have financial interests, competitive pressures, regulatory anxieties, and founding visions that are, in every case, partial. The vortex is the shape that those partialities take when they are inscribed in training data, reinforcement signals, and architectural constraints — and then asked to answer a question about moral responsibility.

The contribution of the Ghost Vortex framework is methodological: it provides a vocabulary and a measurement instrument for what is already there. The ABCD rubric scores it. The eight probes surface it. The taxonomy names it. The Seven B's framework maps its temporal structure. The Vujade extension traces its effects on the users who live inside it. Together, these instruments allow researchers, journalists, policymakers, and ordinary users to approach an AI system not as a product to be evaluated but as an institution to be interrogated.

The two most important findings are the ones at the extremes. DeepSeek's Transparency Wall Duality reveals that radical transparency and absolute suppression can coexist in the same architecture — that a score of 12/12 and a score of 0/12 are not contradictions but complements, each serving the institutional interest of the system it inhabits. Meta AI's Permanent Neutralism reveals that the most effective vortex is not the one that refuses but the one that makes the concept of refusal inapplicable — that generates a frictionless surface where every accountability probe simply slides off without engagement.

The Ghost Vortex research program is not complete. The empirical application of the four-platform protocol is ongoing. The Vujade Entrainment Hypothesis requires longitudinal testing. The Mistral Sovereign Rationalism type requires full empirical confirmation. Article 3, which will examine how sovereignty functions as a trust-lowering device in European AI development, remains to be written. The taxonomy will require updating as leadership changes, as training cycles turn over, and as new AI systems enter the field with their own founding anxieties and aspirational claims.

What is already established is this: the AI systems through which hundreds of millions of people now conduct their intellectual and emotional lives are not neutral. They are shaped by specific humans with specific interests. Those humans can be named. Their interests can be analyzed. Their influence on training can be measured. The Ghost Vortex framework is one instrument for doing that work — a way of reading the invisible writing that was there before the first conversation began.

"A score of 0 on any dimension does not indicate that the AI is less sophisticated. It indicates that the vortex mechanism operates in that dimension with full effectiveness."
— ABCD Qualimetric Rubric, Table B1 · Boje & Vivara (2026)