Representational alignment yields generalizable safety in language models

• URL: http://arxiv.org/abs/2609.04022v1
• Date: Thu, 03 Sep 2026 16:00:16 GMT
• Title: Representational alignment yields generalizable safety in language models
• Abstract summary: We show that prototype-based categorization contributes to behavioral adaptability.<n>We develop representational similarity optimization, which aligns latent representations in large language models with the categorization expressed in human moral judgements.<n>Our findings provide functional support for the view that prototype-based categorization contributes to behavioral adaptability.
• Score: 34.46128626930711
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Aligning large language models (LLMs) is essential for their safe deployment. Current alignment methods mainly optimize observable responses, yet models remain vulnerable when the same harmful intent is recast in unfamiliar or adversarial forms that humans can easily recognize. Prototype theory offers an account of this adaptability. Human concepts are represented around central cases, and new instances are categorized according to their graded typicality relative to these prototypes. Here we show that such categorization of moral concepts is weakly preserved in current LLMs. Across 23 LLMs, models often failed to distinguish opposed moral categories or preserve fine-grained typicality within each category. These deficits persist across parameter sizes and alignment stages. We developed representational similarity optimization, which directly aligns the latent representations in LLMs with the categorization expressed in human moral judgements, without supervising generated responses. In matched experiments using the same 251,334 moral annotations, standard behavioral alignment learned the intended moral judgements at the response level while leaving the categorization structure largely unchanged and increasing vulnerability across adversarial evaluations. Reorganizing moral categorization produced more modest gains in explicit judgements but consistently improved adversarial robustness across model scales on diverse benchmarks and attack strategies. Our findings provide functional support for the view that prototype-based categorization contributes to behavioral adaptability. They also show that transferring this representational principle to LLMs yields generalizable safety under adversarial conditions.

Related papers

• CV-DCLR: Causal-Visual Dynamic Label Refinement for Robust Zero-Shot Learning [25.21342780360303]
  We propose the Causal-Visual Dynamic Label Refinement framework.<n>It recalibrates visual-semantic associations via a Dual-Stream Mutual Correction Mechanism.<n>Experiments on the CUB, SUN, and AWA2 benchmarks demonstrate that CV-DCLR significantly outperforms state-of-the-art methods in high-ambiguity scenarios.
  arXiv  Detail & Related papers  (2026-07-01T12:45:35Z)
• To Compare, or Not to Compare: On Methodological Practices in Evaluating Social Bias [61.40435736418359]
  We introduce a unified framework that standardizes heterogeneous benchmarks to contrast isolated demographic assessments with forced-choice comparative settings.<n>We show that comparative settings act as aggressive catalysts for latent discrimination, a shift primarily driven by underspecified contexts.<n>Finally, we demonstrate that this comparative prejudice is a generalized phenomenon that scales positively with model size.
  arXiv  Detail & Related papers  (2026-06-23T13:53:50Z)
• Does Reasoning Preserve Alignment? On the Trustworthiness of Large Reasoning Models [55.788110316999166]
  We compare reasoning models produced via supervised fine-tuning, RL-based post-training, and distillation against instruction-tuned baselines.<n>We observe that reasoning models often improve on reasoning benchmarks but exhibit alignment regressions.<n>These regressions are consistent with behavioral drift from the instruction-tuned baseline, measured by KL divergence.
  arXiv  Detail & Related papers  (2026-06-09T16:14:27Z)
• From Holistic Evaluation to Structured Criteria: Rubrics Across the Evolving LLM Landscape [79.30826980815927]
  rubrics are explicit criteria sets that transform complex quality judgments into structured and actionable standards.<n>We systematically organize existing rubric designs, examine their construction and optimization, and analyze their role across evaluation and training.
  arXiv  Detail & Related papers  (2026-06-07T13:34:55Z)
• Mechanistic Origin of Moral Indifference in Language Models [17.89411060814224]
  Existing behavioral alignment techniques for Large Language Models (LLMs) often neglect the discrepancy between surface compliance and internal unaligned representations.<n>We verify and remedy this indifference in LLMs' latent representations, utilizing 251k moral vectors constructed upon Prototype Theory and the Social-Chemistry-101 dataset.<n>We then employ Sparse Autoencoders on Qwen3-8B, isolate mono-semantic moral features, and targetedly reconstruct their topological relationships to align with ground-truth moral vectors.
  arXiv  Detail & Related papers  (2026-03-16T17:59:17Z)
• Pressure Reveals Character: Behavioural Alignment Evaluation at Depth [3.634215320925722]
  We introduce an alignment benchmark spanning 904 scenarios across six categories -- Honesty, Safety, Non-Manipulation, Robustness, Corrigibility, and Scheming.<n>Our scenarios place models under conflicting instructions, simulated tool access, and multi-turn escalation to reveal behavioural tendencies that single-turn evaluations miss.<n>We find that even top-performing models exhibit gaps in specific categories, while the majority of models show consistent weaknesses.
  arXiv  Detail & Related papers  (2026-02-24T11:52:17Z)
• Probabilistic Aggregation and Targeted Embedding Optimization for Collective Moral Reasoning in Large Language Models [14.425718737962102]
  We propose a framework that synthesizes multiple LLMs' moral judgments into a collectively formulated moral judgment.<n>Our aggregation mechanism fuses continuous moral acceptability scores (beyond binary labels) into a collective probability.<n>Experiments on a large-scale social moral dilemma dataset show our approach builds robust consensus and improves individual model fidelity.
  arXiv  Detail & Related papers  (2025-06-17T15:22:21Z)
• Estimating Commonsense Plausibility through Semantic Shifts [66.06254418551737]
  We propose ComPaSS, a novel discriminative framework that quantifies commonsense plausibility by measuring semantic shifts.<n> Evaluations on two types of fine-grained commonsense plausibility estimation tasks show that ComPaSS consistently outperforms baselines.
  arXiv  Detail & Related papers  (2025-02-19T06:31:06Z)
• Fundamental Tradeoffs between Invariance and Sensitivity to Adversarial Perturbations [65.05561023880351]
  Adversarial examples are malicious inputs crafted to induce misclassification. This paper studies a complementary failure mode, invariance-based adversarial examples. We show that defenses against sensitivity-based attacks actively harm a model's accuracy on invariance-based attacks.
  arXiv  Detail & Related papers  (2020-02-11T18:50:23Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.