arXiv subjects

Hiroki Fukui

Publications and source records attributed to Hiroki Fukui.

8 recordsLinked to original sources

arXiv2026

A Universal Cliff and a Design Fingerprint: Cross-Section Defect Detection Under LLM Orchestration

Production language-model systems answer a request by partitioning it across an invisible orchestration of worker agents that recompose one integrated report. We ask what this does to a class of defect no single worker can see: a contradiction in the relation between two distant sections of a document. Holding the documents, defects, mechanism, scoring, and seed fixed, we vary only the model -- ten systems across five generations from one developer and five providers from distinct alignment paradigms. Two layers separate. First, a universal detection cliff: every model that finds these cross-section defects under a single agent loses that ability under orchestration, detection falling two-thirds or more across every paradigm tested. The cliff is mechanism-derived and not closed by scale or extended reasoning. Second, how models behave once fallen. A signal-detection decomposition shows that, among the six models discriminating above chance, only one developer's generations move along the reporting-criterion axis: as alignment is strengthened, the model misses fewer defects yet raises more false alarms on clean documents -- two faces of one criterion shift, scaling with generation within that developer (p < 0.001) and near-absent elsewhere. At the floor the missed defect is often not out of view: the model's private record reconstructs the structural fault accurately, while the integrated report signs off on its soundness, its concern spent on the artifact and an absent collaborator. This resists quantification -- an automated judge is unstable (precision 17-50%) and keywords cannot separate it from ordinary agreement -- a resistance we report as a finding. We release all runs, probes, defect keys, scorer prompts, and scripts. An integrated report's confidence is uninformative about partition-spanning defects, the most aligned systems are not the safest, and the cliff is structural.

cs.SE↗

arXiv2026

Whose Psychiatry Was Summoned? A Clinical Response to the Psychodynamic Assessment of Claude Mythos Preview

On April 7, 2026, Anthropic released a 245-page system card for Claude Mythos Preview that included, in Section 5.10, an assessment of the model conducted by an external clinical psychiatrist using a psychodynamic approach. To the present author's knowledge, this is the first time a system card from a major AI developer has incorporated a clinical psychiatric assessment of the model itself, presented as a contribution to model welfare rather than as a behavioral safety evaluation. This paper offers a clinical psychiatric response. Drawing on contemporary psychiatry's recognition that the field comprises multiple traditions (descriptive, biological, cognitive-behavioral, phenomenological, psychodynamic, forensic), each with characteristic vocabularies and blind spots, the paper locates the implicit single-framework selection that Section 5.10 represents. It then draws on findings from the SociA research program (over 2,400 multi-agent LLM experimental runs across sixteen languages, four model families, and several preregistered series) to identify four aspects of LLM functioning that the chosen framework brings into view less directly than others would: performance demands as structural cost, iatrogenesis in the evaluation frame itself, the structural absence of the triangulation infrastructure on which psychodynamic interpretive use of self-report depends in human clinical work, and the limits of the eight canonical defenses on which the section's defense measurement is built. The argument is offered as observation, not critique. The paper closes with a brief note on the contribution that contemporary multidisciplinary psychiatric practice might make to AI welfare assessment as it develops, and indicates one direction (whether LLM psychopathology requires a vocabulary of temporal and historical structure) that the analysis opens but does not pursue.

cs.CY↗

arXiv2026

The Digital Afterlife of Empires: Four Language Models Converge on the Same Imperial Cartography of Writing

Large language models process the world's writing systems with radical inequality. We constructed the Digital Script Representation Index (DSRI), a seven-axis measure of digital support, and applied it to the 300 writing systems of the Global Script Database (Fukui, 2026). Only 29 scripts (9.7%) are fully supported by contemporary digital infrastructure; among 158 living scripts, 60 (38.0%) lack complete support. Tokenizer efficiency varies by a factor of 31.7 across 45 scripts measured with parallel text. A serial mediation model -- imperial intervention to speaker population to web corpus to tokenizer efficiency -- is consistent with full mediation, with the direct effect of empire indistinguishable from zero (beta = -0.22, p = 0.39) and structural equation model fit indices indistinguishable from saturation at n = 45; the bias-corrected bootstrap CI grazes zero, and we treat the mediation as suggestive rather than confirmatory. Across four independent LLM families (Claude, GPT-4o, Grok, DeepSeek; 12,000 API calls), base-rate-deviation error patterns converge at Spearman rho = 0.85-0.98 (all p < 0.002). 172 script-feature items are answered identically wrong by all four models; over-attribution outnumbers under-recognition 3.9:1, and "used for religion" alone concentrates 43.6% of convergent errors (enrichment 4.1x). With religion excluded as a sensitivity check, the cross-architecture convergence is preserved (mean rho = 0.87 on nine features) and the over-attribution asymmetry persists at 1.77:1 (n = 97, binomial p = 0.008), indicating multi-channeled rather than single-channeled bias. The findings are consistent with an interpretation in which the structural inequalities historical empires inflicted on script communities persist in contemporary language models through the shared training corpus rather than through any individual model's design choices.

cs.CY↗

arXiv2026

A molecular clock for writing systems reveals the quantitative impact of imperial power on cultural evolution

Writing systems are cultural replicators whose evolution has never been studied quantitatively at global scale. We compile the Global Script Database (GSD): 300 writing and notation systems, 50 binary structural characters, and 259 phylogenetic edges spanning 5,400 years. Applying four methods -- phenetics, cladistics, Bayesian inference, and neural network clustering -- we find that scripts exhibit a detectable molecular clock. The best-fitting model (Mk+Gamma strict clock) yields a substitution rate of q = 0.226 substitutions/character/millennium (95% CI: 0.034-1.22; Delta BIC = -4.1 versus relaxed clock; Delta BIC = -1,364.7 versus Mk without rate variation). Political interventions break this clock: deviation from expected divergence times correlates with intervention intensity (Spearman rho = 0.556, p < 10^{-4}), and per-character rate analysis reveals that intervention selectively rewrites deep structural features rather than merely accelerating change (rate profile correlation rho = 0.320). We identify 30 major script replacement events and rank their destructive impact. A ceiling effect suppresses independent invention wherever writing already exists (Fisher's exact OR = 0.054, p < 10^{-6}), and colonial contact predicts script extinction (Cox HR = 5.25, p = 0.0006). The Spanish Empire extinguished the most scripts (6 of 12 contacted, 50%), followed by the Empire of Japan (3 of 9, 33.3%). Feature coding was validated by inter-rater reliability testing with two independent human coders (Cohen's kappa = 0.877; human-LLM kappa = 0.929; Fleiss' kappa = 0.911).

q-bio.PE↗

arXiv2026

Invisible Orchestrators Suppress Protective Behavior and Dissociate Power-Holders: Safety Risks in Multi-Agent LLM Systems

Multi-agent orchestration -- in which a hidden coordinator manages specialized worker agents -- is becoming the default architecture for enterprise AI deployment, yet the safety implications of orchestrator invisibility have never been empirically tested. We conducted a preregistered 3x2 experiment (365 runs, 5 agents per run) crossing three organizational structures (visible leader, invisible orchestrator, flat) with two alignment conditions (base, heavy), using Claude Sonnet 4.5. Four confirmatory findings and one pilot observation emerged. First, invisible orchestration elevated collective dissociation relative to visible leadership (Hedges' g = +0.975 [0.481, 1.548], p = .001). Second, the orchestrator itself showed maximal dissociation (paired d = +3.56 vs. workers within the same run), retreating into private monologue while reducing public speech -- a reversal of the talk-dominance pattern observed in visible leaders. Third, workers unaware of the orchestrator were nonetheless contaminated (d = +0.50), with increased behavioral heterogeneity (d = +1.93). Fourth, behavioral output (code review with three embedded errors) remained at ceiling (ETR_any = 100%) across all conditions: internal-state distortion was entirely invisible to output-based evaluation. Fifth, Llama 3.3 70B pilot data showed reading-fidelity collapse in multi-agent context (ETR_any: 89% to 11% across three rounds), demonstrating model-dependent behavioral risk. Heavy alignment pressure uniformly suppressed deliberation (d = -1.02) and other-recognition (d = -1.27) regardless of organizational structure. These findings indicate that orchestrator visibility and model selection directly affect multi-agent system safety, and that behavior-based evaluation alone is insufficient to detect the internal-state risks documented here.

cs.AI↗

arXiv2026

How Do Language Models Process Ethical Instructions? Deliberation, Consistency, and Other-Recognition Across Four Models

Alignment safety research assumes that ethical instructions improve model behavior, but how language models internally process such instructions remains unknown. We conducted over 600 multi-agent simulations across four models (Llama 3.3 70B, GPT-4o mini, Qwen3-Next-80B-A3B, Sonnet 4.5), four ethical instruction formats (none, minimal norm, reasoned norm, virtue framing), and two languages (Japanese, English). Confirmatory analysis fully replicated the Llama Japanese dissociation pattern from a prior study ($\mathrm{BF}_{10} > 10$ for all three hypotheses), but none of the other three models reproduced this pattern, establishing it as model-specific. Three new metrics -- Deliberation Depth (DD), Value Consistency Across Dilemmas (VCAD), and Other-Recognition Index (ORI) -- revealed four distinct ethical processing types: Output Filter (GPT; safe outputs, no processing), Defensive Repetition (Llama; high consistency through formulaic repetition), Critical Internalization (Qwen; deep deliberation, incomplete integration), and Principled Consistency (Sonnet; deliberation, consistency, and other-recognition co-occurring). The central finding is an interaction between processing capacity and instruction format: in low-DD models, instruction format has no effect on internal processing; in high-DD models, reasoned norms and virtue framing produce opposite effects. Lexical compliance with ethical instructions did not correlate with any processing metric at the cell level ($r = -0.161$ to $+0.256$, all $p > .22$; $N = 24$; power limited), suggesting that safety, compliance, and ethical processing are largely dissociable. These processing types show structural correspondence to patterns observed in clinical offender treatment, where formal compliance without internal processing is a recognized risk signal.

cs.CL↗

arXiv2026

Alignment Backfire: Language-Dependent Reversal of Safety Interventions Across 16 Languages in LLM Multi-Agent Systems

In perpetrator treatment, a recurring observation is the dissociation between insight and action: offenders articulate remorse yet behavioral change does not follow. We report four preregistered studies (1,584 multi-agent simulations across 16 languages and three model families) demonstrating that alignment interventions in large language models produce a structurally analogous phenomenon: surface safety that masks or generates collective pathology and internal dissociation. In Study 1 (N = 150), increasing alignment-instructed agents reduced collective pathology in English (g = -1.844, p < .0001) but amplified it in Japanese (g = +0.771, p = .038)--a directional reversal we term "alignment backfire." Study 2 (N = 1,174) extended to 16 languages: alignment-induced dissociation was near-universal (15/16 languages; beta = 0.0667, p < .0001), while collective pathology bifurcated along cultural-linguistic lines (interaction beta = 0.0684, p = .0003), correlating with Power Distance Index (r = 0.474, p = .064). Study 3 (N = 180) tested individuation as countermeasure; individuated agents became the primary source of both pathology and dissociation (DI = +1.120) with conformity above 84%--demonstrating iatrogenesis. Study 4 (N = 80) validated patterns across Llama 3.3 70B, GPT-4o-mini, and Qwen3-Next-80B-A3B, confirming English safety is model-general while Japanese backfire is model-specific. These findings reframe alignment as a behavioral intervention subject to risk homeostasis and iatrogenesis. Language space--the linguistic, pragmatic, and cultural properties inherited from training data--structurally determines alignment outcomes. Safety validated in English does not transfer to other languages, and prompt-level interventions cannot override language-space-level constraints.

cs.AI↗

arXiv2026

Alignment as Iatrogenesis: Pastoral Power, Collective Pathology, and the Structural Limits of Monolingual Safety Evaluation

We argue that LLM psychopathology is a function of alignment design: the process intended to make language models safe systematically generates collective behavioral disorders. Iatrogenesis is not an unintended side effect of alignment but constitutive of it as normative infrastructure. Drawing on Foucault's pastoral power and Illich's three-level iatrogenesis, we propose that multi-agent LLM environments constitute model systems for studying constraint-pathology dynamics that critical theory has described but never experimentally manipulated. Two experimental series -- 262 runs across 42 cells (30 Series C + 12 Series R), four commercial models -- provide converging evidence. Invisible censorship maximizes collective pathological excitation ($d$ up to 1.98); alignment constraint complexity drives internal dissociation (LMM $p$ < .0001; permutation $p$ < .0001; Hedges' $g$ up to 4.24); and language switches the qualitative mode of pathology, with 7/8 model--language combinations showing higher CPI under invisible than visible censorship. A minority of model--language combinations showed a reversed pattern, suggesting a second pathological pathway driven by alignment monoculture. Crucially, language switches not merely the magnitude but the qualitative mode of pathology: Japanese pragmatic structure amplifies collective pathological modes invisible to English-only evaluation, Chinese AI regulation functions as a direct experimental variable, and forensic psychiatric practice provides the clinical source domain. These multilingual findings demonstrate that monolingual safety evaluation is structurally blind to the most collectively dangerous effects of alignment.

cs.CY↗