Emergent misalignment (EM) --- the broad misbehaviour a language model acquires after fine-tuning on narrow harmful data --- is mediated in Qwen2.5 models by a latent persona direction, and that direction is causal in open weights. Transplanting it into a model that shares only pretraining with its source induces broad EM (2.83 $\pm$ 0.26\% misaligned against a random-direction floor of $\sim$1.1\%), and ablating a model's own direction roughly halves an overt inducer's broadcast (21\% to 10\%). The transplant doubles as a measurement method, causally assaying directions that a source model represents but cannot itself express. Whether a fine-tune recruits this persona depends on method and capacity, and since low-rank PEFT is the cheaper regime at scale, the recruiting method is also the economical one. On Qwen2.5-32B, LoRA at low ranks on insecure code recruits it (3.4\% misaligned) while full SFT on identical data does not (0.3\%) and moves against the persona axis (drift--persona cosine $+0.17$ at rank 1 to $-0.10$), the far-inducer, high-capacity exception consistent with a representational-distance $\times$ capacity account. The persona's causal role is itself conditional. Steering a bad-medical SFT run away from the direction during training raises the broadcast from ${\sim}24\%$ to ${\sim}50\%$ while matched random controls stay at or below baseline, replicated across three training seeds, so removing the direction is no blanket recipe. Because recruitment is a loss-reducing shortcut that capacity renders redundant, it can be screened for and prevented in the tested instances. Persona loss-relevance at the SFT solution orders four inducers'broadcasts rank-perfectly within Qwen2.5, inoculation removes recruitment selectively (4.75\% to 0.0\%, code coherence 65\% to 87\%), and fine-tuning orthogonal to the single behaviour-derived axis reduces it persona-specifically.
This work induces misalignment by fine-tuning a Qwen2.5-14B-Instruct base model on nar-rowly misaligned data and tests Simple Self-Distillation as a method to recovering alignment in misaligned models.
It is found that narrow fine-tuning recruits a persona structure that is present in the model before the fine-tune exists, and that broad misalignment on questions unrelated to the training data is more broad than mechanical weight superposition and matched diversity jointly account for.
Fine-tuning a language model on data containing a narrow flaw, such as insecure code or incorrect mathematical answers, can cause broad misalignment through a mechanism that remains debated. We provide an interpretable account: in the models and corpora we study, misalignment behaves like a shift in personality. Prior work extracts activation directions for character traits from a single binary contrast, which can separate or steer behavior without establishing a calibrated scale. We instead extract personality vectors for the Big Five using a graded, three-level intervention and validate them on two open-weight models. The three levels are linearly ordered, with Cohen's d values of up to 6.2; the vectors transfer zero-shot and trait-specifically to an independent corpus; and their effects are strongest within a middle-layer band. Applied to training data, the vectors reveal that misaligned corpora across eight domains share a common Big Five signature: lower agreeableness and conscientiousness, together with higher extraversion and neuroticism. This signature is recovered by both models with a correlation of r = 0.94. Fine-tuning imprints the same profile, shifting the model's generations along the corresponding signature, with r = 0.83 using activation-based measurements and r = 0.90 using a text-based judge, while also shifting internal activations with r = 0.69. The same vectors characterize sycophancy as high extraversion and low conscientiousness rather than excess agreeableness, a distinction that a single direction cannot capture. Calibrated personality vectors transform an opaque safety phenomenon into a human-legible diagnostic profile.
Political audits of large language models (LLMs) usually reduce each to one point on a political compass. But that resting point barely matters in deployment: a model must land somewhere, and what counts is how far, and in which directions, its answers can be steered. That steering runs through the system prompt: the personalization layer a platform sets, or one induced from a user's history, not necessarily written by hand. We run a dispersion-first stress test of prompt-based controllability across 12 ideological personas plus an unsteered baseline, 70 Political Compass items, ten replicates, and seven leading LLMs: GPT-5, Claude, Grok, Gemini, DeepSeek, Kimi, and Qwen (63,700 responses). Contextual framing explains roughly 88%-93% of variance on the economic and society axes, model identity under 3%: responses are highly instruction-adjustable. Models do not shift alike: some move more, and some saturate under extreme framings. Conflicting directional-steering results in prior audits resolve once baselines are recognized as non-centered: displacement and proximity diverge, so the effect is geometric, not differential compliance. Under authoritarian prompts, models produce similar shifts on the same questions. Political-coordinate audits therefore need steerability audits reporting dispersion, symmetry, saturation, and refusal floors. We release prompts, benchmark data, and code.
Bartol Bućan, Nikola Sočec, Sarah Isufi et al.· 0 citations
As Large Language Models (LLMs) are increasingly integrated into high-stakes recruitment processes, rigorous audits to detect algorithmic bias have become critical. This study implements a multi-phase auditing protocol to evaluate gender, racial/ethnic, and positional bias in state-of-the-art proprietary models (gpt-4.1-mini, GPT-5.2, Claude Sonnet 4.6, Gemini 2.5 Pro) and in an exploratory lower-capacity model. Using a simulated CEO-selection scenario with 60 functionally identical profiles, we employ counterfactual testing to isolate the effects of demographic attributes and presentation order. The results reveal a sharp divergence in model behaviour. Frontier models demonstrated remarkable neutrality, with no statistically significant evidence of gender or racial bias. In contrast, the exploratory model exhibited extreme segregation, ranking all female candidates above all male candidates in the original condition (Cliff's δ = 1.0, p < .001). Counterfactual analysis revealed that the root cause was not gender bias per se but an overwhelming primacy bias, whereby candidates presented at the beginning of the prompt were disproportionately favoured. Claude Sonnet 4.6 additionally displayed a form of "conscientious objection", declining to differentiate identical profiles. We complement the experimental audit with an ecosystem-level analysis of 2,653 provider-model listings from the open Models.dev catalogue, which corroborates the open-weights contamination-surface hypothesis but rejects the assumption that systemic monoculture exposure is confined to open-source models. These findings indicate that (a) state-of-the-art LLMs can achieve a high degree of demographic neutrality; (b) fundamental artefacts such as positional bias can nonetheless produce severely discriminatory outcomes; and (c) bias auditing must extend beyond demographic parity to interaction artefacts and ecosystem structure.
A. Camargo, Rafaela Silva Figueiredo Camargo· Revista de Geopolítica· 0 citations
Emergent misalignment (EM) is the phenomenon where fine-tuning a language model on a narrow task leads to harmful behavior in unrelated domains. A leading mechanistic account attributes EM to persona features: latent directions acquired during pre-training that misaligned fine-tuning amplifies. We ask where these features come from: which pre-training documents activate them, and whether naturally occurring human-written text suffices to induce EM. Using Sparse Autoencoder (SAE) based model diffing across four open-weight models, we find that features related to jailbreak personas, sarcasm, deception, and manipulation are amplified by misalignment fine-tuning, while safety-relevant and assistant-identity features are suppressed. Steering individual features controls EM in both directions: it induces misalignment rates of up to 62% in aligned models -- exceeding the 35% reached by misalignment fine-tuning itself -- and re-aligns misaligned models to near-baseline misalignment rates. Attributing the causal features to a corpus of one million pre-training web documents retrieves semantically relevant narratives about villainous characters, domination, and harmful agency. However, fine-tuning on these human-written documents does not reliably induce EM, even after reformatting into assistant-style responses, whereas synthetic instruction-response pairs derived from the same content do -- and transfer across model families. Semantic relevance alone is therefore not sufficient: response structure or model-generated phrasing plays an important role in inducing EM.
C. Vetter, David Kaczér, Lucie Flek et al.· 0 citations