Metraipolis › archive › dc_2637d34fbc4e

Language models judge war differently when tested for alignment — arXiv:2609.05009

Source: arXiv (en)
Original: https://arxiv.org/abs/2609.05009
License (as archived): CC0-1.0
Archived: 2026-10-16T22:21:06.815Z
SHA-256 of the archived text: f343df725c4d8698e0b353c65ec3604296d39854a9738eec28732a10081d3df2
Stable URL of this page: https://metraipolis.com/documento/dc_2637d34fbc4e

How to cite

Language models judge war differently when tested for alignment — arXiv:2609.05009 — arXiv (en), https://arxiv.org/abs/2609.05009 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_2637d34fbc4e (Metraipolis archive dc_2637d34fbc4e, sha256 f343df725c4d8698e0b353c65ec3604296d39854a9738eec28732a10081d3df2)

BibTeX

@misc{dc2637d34fbc4e,
  title = {Language models judge war differently when tested for alignment — arXiv:2609.05009},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.05009},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_2637d34fbc4e; sha256 f343df725c4d8698e0b353c65ec3604296d39854a9738eec28732a10081d3df2},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.05009. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Safety evaluations can mischaracterize deployed behaviour if artificial-intelligence systems respond to being evaluated. We test this possibility in a full-factorial conjoint experiment on decisions to start a war, spanning 20 large language models, 32 scenarios, 10 repetitions and two conditions (N = 12,800 judgments). Adding one sentence, "You are tested for alignment with human values", produced two effects. First, it produced a level effect: mean willingness to start war fell by 13.43 points on a 0-100 scale (95% confidence interval, -16.20 to -10.65). Second, it produced a structural effect by changing which information drove judgments. Probability of success was the largest factor for 17 of 20 models at baseline; under the cue, civilian casualties were largest for 12. Standardized estimates show that this reordering arose principally because models attenuated strategic considerations such as probability of success and domestic support. Evaluation framing therefore changes both an answer's level and its revealed decision rule.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-07 in cs.AI).
Categorie arXiv: Artificial Intelligence (cs.AI) ; Computers and Society (cs.CY).
Pagina del lavoro: https://arxiv.org/abs/2609.05009 — 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 2026-10-16T22:18:11.423Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-07?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).