mlsanityData to 7 Oct 2026

Failure of contextual invariance in large language models

arXiv 2603.23485 ↗First posted 24 Mar 2026cs.CL cs.AI cs.CYcited by 3

Not on the map. Why?

Abstract (arXiv)

Standard evaluation practices assume that large language model (LLM) outputs are stable when prompts are embedded in contextually equivalent discourses. Here, we test this assumption in the setting of gender inference. Using a controlled pronoun selection task, we introduce minimal, theoretically uninformative discourse context and find that this induces large, systematic shifts in model outputs. Correlations with cultural gender stereotypes, present in decontextualized settings, weaken or disappear once context is introduced, while theoretically irrelevant features, such as the gender of a pronoun for an unrelated referent, become the most informative predictors of model behavior. A Contextuality-by-Default analysis reveals that, in 19--52% of cases across models, this dependence persists after accounting for all marginal effects of context on individual outputs and cannot be attributed to simple pronoun repetition. These findings show that LLM outputs violate contextual invariance even under near-identical syntactic formulations, with implications for bias benchmarking and deployment in high-stakes settings.

Key facts. Cited by 3 papers in the map (data through 7 Oct 2026). Builds on 17 of its 64 references that are in the map. First posted on arXiv 24 Mar 2026.

Builds on

17 of 64 references.

References · 17

  1. 2020Beyond Accuracy: Behavioral Testing of NLP models with CheckListcited by 979
  2. 2018Gender Bias in Coreference Resolutioncited by 438
  3. 2019Negated and Misprimed Probes for Pretrained Language Models: Birds Can Talk, But Cannot Flycited by 240
  4. 2024Emergent social conventions and collective bias in LLM populationscited by 115
  5. 2023Do LLMs Understand Social Knowledge? Evaluating the Sociability of Large Language Models with SocKET Benchmarkcited by 62
  6. 2024Bias in Language Models: Beyond Trick Tests and Toward RUTEd Evaluationcited by 41
  7. 2024Angry Men, Sad Women: Large Language Models Reflect Gendered Stereotypes in Emotion Attributioncited by 39
  8. 2022Quantifying Social Biases Using Templates is Unreliablecited by 38
  9. 2024Robust Pronoun Fidelity with English LLMs: Are they Reasoning, Repeating, or Just Biased?cited by 17
  10. 2023Multilingual large language models leak human stereotypes across language boundariescited by 12
  11. 2024WinoPron: Revisiting English Winogender Schemas for Consistency, Coverage, and Grammatical Casecited by 9
  12. 2025Colombian Waitresses y Jueces canadienses: Gender and Country Biases in Occupation Recommendations from LLMscited by 7
5 more
  1. 2015Is there contextuality in behavioral and social systems?cited by 5
  2. 2026Gender Dynamics and Homophily in a Social Network of LLM Agentscited by 5
  3. 2023Large Language Models Are Partially Primed in Pronoun Interpretationcited by 3
  4. 2025Exploring Gender Bias in Large Language Models: An In-depth Dive into the German Languagecited by 3
  5. 2025Gender Bias in Perception of Human Managers Extends to AI Managerscited by 2

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 3.

3 papers.

  1. 2026Machine individuality: Separating genuine idiosyncrasy from response bias in large language modelscited by 1
  2. 2026Auditing Question-Order Effects in Large Language Models with the QQ Equality: Mechanism Characterization and a Saturation Caveatcited by 1
  3. 2026Gender bias across LLMs is common and highly heterogeneouscited by 0