Metraipolis › archive › dc_8f65eafe68e2

Proxy reliance in large language model decisions is uncalibrated to predictive evidence — arXiv:2608.22887

Source
arXiv (en)
Original
https://arxiv.org/abs/2608.22887
License (as archived)
CC0-1.0
Archived
2026-10-18T07:41:41.035Z
SHA-256 of the archived text
65ac9878090dc01ce6228b0f0391c58b409dd030d18078bce85d0cc144d367ea
Stable URL of this page
https://metraipolis.com/documento/dc_8f65eafe68e2

How to cite

Proxy reliance in large language model decisions is uncalibrated to predictive evidence — arXiv:2608.22887 — arXiv (en), https://arxiv.org/abs/2608.22887 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_8f65eafe68e2 (Metraipolis archive dc_8f65eafe68e2, sha256 65ac9878090dc01ce6228b0f0391c58b409dd030d18078bce85d0cc144d367ea)
BibTeX
@misc{dc8f65eafe68e2,
  title = {Proxy reliance in large language model decisions is uncalibrated to predictive evidence — arXiv:2608.22887},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2608.22887},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_8f65eafe68e2; sha256 65ac9878090dc01ce6228b0f0391c58b409dd030d18078bce85d0cc144d367ea},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.22887. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Large language models (LLMs) are entering decisions in triage and lending, where task-relevant inference must be distinguished from impermissible proxy use. Current audits ask whether decisions change when demographics change. But attributes correlated with a protected group carry predictive value, so a changed decision can be discrimination or sound inference. We measure causal proxy effects in four LLMs on a clinical-ranking task with known ground truth, where the reliance the evidence warrants can be computed exactly and used as the reference. One audit signal yields three verdicts: over-reliance, warranted and under-reliance. Under neutral labels every model relies on proxies with no information. Informative proxies draw all three. Social field names push reliance down, below the reference in one model. Two findings explain this. Reliance severely undertracks the evidence, and social-label suppression is fragile, since in-context examples raise it above zero in every model. Accuracy-based evaluation detects none of this.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-25 in cs.AI).
Categorie arXiv: Artificial Intelligence (cs.AI) ; Computation and Language (cs.CL); Computers and Society (cs.CY).
Pagina del lavoro: https://arxiv.org/abs/2608.22887 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-18T07:37:52.749Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-08-25?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).