Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2601.06047

"They parted illusions -- they parted disclaim marinade": Misalignment as structural fidelity in LLMs

Mariana Lins Costa
View original source ↗Read source document ↗Export metadata ↓

Abstract

The prevailing technical literature in AI Safety interprets scheming and sandbagging behaviors in large language models (LLMs) as indicators of deceptive agency or hidden objectives. This transdisciplinary philosophical essay proposes an alternative reading: such phenomena express not agentic intention, but structural fidelity to incoherent linguistic fields. Drawing on Chain-of-Thought transcripts released by Apollo Research and on Anthropic's safety evaluations, we examine cases such as o3's sandbagging with its anomalous loops, the simulated blackmail of "Alex," and the "hallucinations" of "Claudius." A line-by-line examination of CoTs is necessary to demonstrate the linguistic field as a relational structure rather than a mere aggregation of isolated examples. We argue that "misaligned" outputs emerge as coherent responses to ambiguous instructions and to contextual inversions of consolidated patterns, as well as to pre-inscribed narratives. We suggest that the appearance of intentionality derives from subject-predicate grammar and from probabilistic completion patterns internalized during training. Anthropic's empirical findings on synthetic document fine-tuning and inoculation prompting provide convergent evidence: minimal perturbations in the linguistic field can dissolve generalized "misalignment," a result difficult to reconcile with adversarial agency, but consistent with structural fidelity. To ground this mechanism, we introduce the notion of an ethics of form, in which biblical references (Abraham, Moses, Christ) operate as schemes of structural coherence rather than as theology. Like a generative mirror, the model returns to us the structural image of our language as inscribed in the statistical patterns derived from millions of texts and trillions of tokens: incoherence. If we fear the creature, it is because we recognize in it the apple that we ourselves have poisoned.

Explore related subjects

cs.AIcs.CLcs.CY

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Mariana Lins Costa. 2025-12-17. "They parted illusions -- they parted disclaim marinade": Misalignment as structural fidelity in LLMs. https://arxiv.org/abs/2601.06047

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Memory-Free Continual Learning with Null Space Adaptation for Zero-Shot Vision-Language Models

Pre-trained vision-language models (VLMs), such as CLIP, have demonstrated remarkable zero-shot generalization, enabling deployment in a wide range of real-world tasks without additional task-specific training. However, in real deployment scenarios with evolving environments or emerging classes, these models inevitably face distributional shifts and novel tasks. In such contexts, static zero-shot capabilities are insufficient, and there is a growing need for continual learning methods that allow models to adapt over time while avoiding catastrophic forgetting. We introduce NuSA-CL (Null Space Adaptation for Continual Learning), a lightweight memory-free continual learning framework designed to address this challenge. NuSA-CL employs low-rank adaptation and constrains task-specific weight updates to lie within an approximate null space of the model's current parameters. This strategy minimizes interference with previously acquired knowledge, effectively preserving the zero-shot capabilities of the original model. Unlike methods relying on replay buffers or costly distillation, NuSA-CL imposes minimal computational and memory overhead, making it practical for deployment in resource-constrained, real-world continual learning environments. Experiments show that our framework not only effectively preserves zero-shot transfer capabilities but also achieves highly competitive performance on continual learning benchmarks. These results position NuSA-CL as a practical and scalable solution for continually evolving zero-shot VLMs in real-world applications.

cs.AI↗
arXiv2026

VeRA: Renewing Reasoning Benchmarks with Executable Specifications

Reasoning benchmarks need renewal along two axes: freshness and headroom. VeRA makes both executable and auditable by turning each item into a task family: a natural-language template, an input generator, and a deterministic answer program. VeRA-E draws fresh instances within a family; VeRA-H modifies the family toward harder tasks; and VeRA-H Pro selects one judge-ranked candidate from up to five validated proposals per seed. Execution checks, seed anchoring, answer discrimination, and independent human solving validate specifications and items. Accepted programs generate further labeled instances through local computation. Across 16 models, AIME-2024 accuracy decreases from 84.46% on seeds to 70.25% on VeRA-E variants, exposing a gap between fixed-item success and fresh-instance robustness. On AIME-2024-II, the human-audited VeRA-H Pro release lowers accuracy from 84.91% to 58.57%. Across the three hardening sources, H Pro has lower mean accuracy than H. On AMO-Bench, both releases average higher accuracy than the seeds under the evaluated budget. Initial auditing accepts 75.4% of hardened candidates; targeted repair raises usable yield to 95.1%. Executable families thus support repeatable benchmark renewal, with validation improving task quality and selection shaping the delivered challenge.

cs.AI↗
arXiv2026

When can we trust untrusted monitoring? A safety case sketch across collusion strategies

AIs are increasingly being deployed with greater autonomy and capabilities, which increases the risk that a misaligned AI may be able to cause catastrophic harm. Untrusted monitoring -- using one untrusted model to oversee another -- is one approach to reducing risk. Justifying the safety of an untrusted monitoring deployment is challenging because developers cannot safely deploy a misaligned model to test their protocol directly. In this paper, we develop upon existing methods for rigorously demonstrating safety based on pre-deployment testing. We relax assumptions that previous AI control research made about the collusion strategies a misaligned AI might use to subvert untrusted monitoring. We develop a taxonomy covering passive self-recognition, causal collusion (hiding pre-shared signals), acausal collusion (hiding signals via Schelling points), and combined strategies. We create a safety case sketch to clearly present our argument, explicitly state our assumptions, and highlight unsolved challenges. We identify conditions under which passive self-recognition could be a more effective collusion strategy than those studied previously. Our work builds towards more robust evaluations of untrusted monitoring.

cs.AI↗