{
  "abstract": "Large language models (LLMs) exhibit us-vs.-them bias: A behavioral asymmetry in which prompts framed around an ingroup (``we''/``us'') receive systematically more positive continuations than matched prompts framed around an outgroup (``they''/``them''). Using Edge Attribution Patching (EAP), we localize this behavior to directed circuits in GPT-2 Small, GPT-2 Large, Llama-2 7B, and Llama-3 8B. Across models, high-attribution edges are sparse and concentrated primarily in middle-to-late layers, while ingroup- and outgroup-related pathways are only partially overlapping and reorganize under fine-tuning. Behavioral comparisons and 30-token generations show that the measured asymmetry extends beyond next-token evaluation. Targeted patching of EAP-ranked edges reduces the bias score more than equal-cardinality random-edge controls, but also substantially degrades performance on CoLA, HellaSwag, and WikiText-2, suggesting that bias-relevant pathways are functionally entangled with broader language capabilities. Overall, our findings characterize us-vs.-them bias as a sparse, context-dependent, and functionally entangled computational phenomenon whose internal organization varies across the models examined.",
  "abstract_shown": true,
  "affiliations": [],
  "article_type": null,
  "arxiv": "2610.10864",
  "author_total": 4,
  "authors": [
    {
      "name": "Tabia Tanzin Prama"
    },
    {
      "name": "Julia Witte Zimmerman"
    },
    {
      "name": "Christopher M. Danforth"
    },
    {
      "name": "Peter Sheridan Dodds"
    }
  ],
  "bibtex": "https://research.draconitic.com/paper/7r40339q8zfmn.bib",
  "cited_by": 0,
  "doi": "10.48550/arxiv.2610.10864",
  "fulltext_url": null,
  "id": "7r40339q8zfmn",
  "issns": [],
  "landing_url": "https://arxiv.org/abs/2610.10864",
  "language": null,
  "licence": "LicenseRef-arXiv-nonexclusive-1.0",
  "licence_label": "arXiv licence",
  "licence_source": "article",
  "licence_url": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
  "metadata_licence": "CC0-1.0",
  "native_subjects": [
    [
      "arxiv",
      "physics.soc-ph"
    ],
    [
      "openalex",
      "3311|Ethics and Social Impacts of AI"
    ]
  ],
  "openalex": "W7221188398",
  "pdf_url": "https://arxiv.org/pdf/2610.10864",
  "pmcid": null,
  "pmid": null,
  "published": "2026-10-07",
  "publisher": null,
  "ris": "https://research.draconitic.com/paper/7r40339q8zfmn.ris",
  "sources": [
    {
      "first_seen": "2026-10-30T12:27:54Z",
      "native_id": "2610.10864",
      "source": "arxiv",
      "url": "https://arxiv.org/abs/2610.10864"
    }
  ],
  "subjects": [
    {
      "field": "physics",
      "origin": "arxiv:physics.soc-ph",
      "subfield": "statistical-nonlinear"
    }
  ],
  "title": "Localized, unstable, and entangled: Exploring `Us vs. Them' bias in Large Language Models",
  "title_text": "Localized, unstable, and entangled: Exploring `Us vs. Them' bias in Large Language Models",
  "type": "preprint",
  "updated_at": "2026-10-30T12:27:54Z",
  "url": "https://research.draconitic.com/paper/7r40339q8zfmn",
  "venue": "arXiv",
  "versions": []
}