mlsanityData to 14 Oct 2026

The Alignment Veto: How Safety Training Suppresses Cultural Knowledge in LLMs

arXiv 2510.13154 ↗First posted 15 Oct 2025cs.CLcited by 3

On the map: LLM alignment / RLHF / preference optimisation → Human values and AI feedback alignment

Abstract (arXiv)

What happens inside a language model when alignment training conflicts with a cultural value it encodes? Across 16 MENA countries, 26 models, and 1.53M human survey responses, we show the answer is suppression, not erasure: at the moment of refusal, a model's internal logit distribution correlates with human survey data more strongly than its freely generated answers. We call this the alignment veto. We distinguish suppression failures (accurate internal distributions blocked at output) from representational bias failures (the encoding itself diverges from human values), and show the two require different interventions. The gate is inequitable: the safety tax reaches 37.6%, with a 19.8% alignment-quality gap between best- and worst-served nations, and native-language prompting widens rather than closes it. Sparse autoencoder analysis corroborated by comparisons across alignment stages identifies a candidate DPO-stage feature mediating suppression in Tulu-3-8B. The gate is real, its costs are unequal, and deciding what it should protect is not a technical question.

Key facts. Cited by 3 papers in the map (data through 14 Oct 2026). Placed in Human values and AI feedback alignment, one of 10 directions in LLM alignment / RLHF / preference optimisation. Builds on 14 of its 27 references that are in the map. First posted on arXiv 15 Oct 2025.

Builds on

14 of 27 references.

Closest ancestors · 1

Same direction, newest first

  1. 2024Investigating Cultural Alignment of Large Language Models20 Feb

Other references · 13

  1. 2022Training language models to follow instructions with human feedbackcited by 15,971
  2. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 8,065
  3. 2017Deep reinforcement learning from human preferencescited by 3,912
  4. 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 1,420
  5. 2024Refusal in Language Models Is Mediated by a Single Directioncited by 1,032
  6. 2023Whose Opinions Do Language Models Reflect?cited by 773
  7. 2023Towards Measuring the Representation of Subjective Global Opinions in Language Modelscited by 437
  8. 2023Assessing Cross-Cultural Alignment between ChatGPT and Human Societies: An Empirical Studycited by 206
  9. 2023Having Beer after Prayer? Measuring Cultural Bias in Large Language Modelscited by 153
  10. 2022Probing Pre-Trained Language Models for Cross-Cultural Differences in Valuescited by 149
  11. 2024The Language Barrier: Dissecting Safety Challenges of LLMs in Multilingual Contextscited by 94
  12. 2024Unintended Impacts of LLM Alignment on Global Representationcited by 66
1 more
  1. 2025LLM Alignment for the Arabs: A Homogenous Culture or Diverse Ones?cited by 6

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 3.

3 papers.

  1. 2026The Shibboleth Effect: Auditing the Cross-Lingual Distributional Skew of Large Language Modelscited by 0
  2. 2026GYROval: A Robust Benchmark for Cultural Value Orientation in Large Language Modelscited by 0
  3. 2026Large-scale factor analysis shows machine intelligence is only partially interpretablecited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Aligned but Flattened: Analyzing the Trade-off between Cultural Alignment and Diversity in LLMs
  2. 2025Randomness, Not Representation: The Unreliability of Evaluating Cultural Alignment in LLMs
  3. 2026Whose Norms? Disentangling Cultural and Personal Alignment in Large Language Models
  4. 2025Alignment and Safety in Large Language Models: Safety Mechanisms, Training Paradigms, and Emerging Challenges
  5. 2024Navigating the Cultural Kaleidoscope: A Hitchhiker's Guide to Sensitivity in Large Language Models
  6. 2026The Limits of Simulated Societies: How Post-Training and Survey Fine-Tuning Erase Cross-Cultural Variance
  7. 2026Scenario-based Probing and Steering Cultural Values in Large Language Models--Extended Version
  8. 2026Are Aligned Large Language Models Still Misaligned?