Metraipolis › archive › dc_a7cee51aed46

Action Verification through Conjecture and Refutation for Language Model Agents — arXiv:2611.02950

Source
arXiv (en)
Original
https://arxiv.org/abs/2611.02950
License (as archived)
CC0-1.0
Archived
2026-11-03T10:33:36.535Z
SHA-256 of the archived text
b771fc85b8fc226030dc8d69dd67a691aeda528f0ba4f9000fdcde80205bad3d
Stable URL of this page
https://metraipolis.com/documento/dc_a7cee51aed46

How to cite

Action Verification through Conjecture and Refutation for Language Model Agents — arXiv:2611.02950 — arXiv (en), https://arxiv.org/abs/2611.02950 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_a7cee51aed46 (Metraipolis archive dc_a7cee51aed46, sha256 b771fc85b8fc226030dc8d69dd67a691aeda528f0ba4f9000fdcde80205bad3d)
BibTeX
@misc{dca7cee51aed46,
  title = {Action Verification through Conjecture and Refutation for Language Model Agents — arXiv:2611.02950},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2611.02950},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_a7cee51aed46; sha256 b771fc85b8fc226030dc8d69dd67a691aeda528f0ba4f9000fdcde80205bad3d},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.02950. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Language model agents increasingly act on users' behalf by manipulating external environments through tools, often from high-level instructions that leave critical task- and environment-specific context unstated. To prevent resulting misaligned actions, recent work verifies proposed actions before execution, largely by establishing whether the conditions supporting them hold. We argue that this confirmation-oriented view leaves a fundamental blind spot: a wrong action often passes every check that asks whether it is right. This motivates a different question for verification: rather than asking what evidence supports an action, can the agent find evidence that refutes it? We pursue this direction with BlackSwan, a refutation-oriented, expect-suspect-test procedure that predicts the intended effect of each state-mutating action, formulates specific and empirically testable hypotheses for how it could be wrong, and tests them using existing observations or additional non-mutating actions, executing the action only if none survives testing. On challenging AppWorld tasks, BlackSwan improves success rate over the strongest baseline by 14.9 and 25.2%p with and without the backbone model's thinking mode, respectively. Even without thinking or in-context demonstrations, it outperforms stronger baselines equipped with them. The gains also transfer beyond AppWorld, with BlackSwan improving WebShop performance by 12.7%p. More broadly, our results suggest that language model agents have an underused capacity to recognize and self-correct misaligned actions, which can be systematically elicited through active conjecture and refutation before execution.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Artificial Intelligence (cs.AI) ; Computation and Language (cs.CL); Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2611.02950 — 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 2026-11-03T10:29:48.612Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).