Newest papers

When LLMs Forget What They Believe: Characterizing State Integrity Collapse in Long-Horizon Dialogues

Yancheng Zhu, Yifan Yuan · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.31314 · cited by 0 works (OpenAlex) · open-access PDF

French

Nous étudions cette dissociation de rappel-conformité dans un dialogue long-horizon et définissons l'intégrité de l'État : un modèle conserve une contrainte déclarée dans la mémoire mais arrête de l'appliquer à des tours ultérieurs. Nous introduisons StateTrack-Bench, une suite contrôlée de conversations couvrant le numérique, le format, le rôle/personne, la mémoire factuelle et les contraintes de règle logique. Sur sept snapshots du modèle de fournisseur et cinq sondes de profondeur de dialogue, une stricte rétention conjointe varie de 62,4% à 100,0%, avec une dégradation concentrée dans des contraintes symboliques et dans un modèle particulièrement fragile. L'audit de rappel aigué

Machine translation (Meta M2M100); the English original follows.

Abstract

Large language models can repeat an instruction while failing to keep obeying it. We study this recall-compliance dissociation in long-horizon dialogue and define State Integrity Collapse: a model preserves a stated constraint in memory but stops applying it to later turns. We introduce StateTrack-Bench, a controlled suite of conversations covering numerical, format, role/persona, factual-memory, and logical-rule constraints. Across seven provider-pinned model snapshots and five dialogue-depth probes, strict joint retention ranges from 62.4% to 100.0%, with degradation concentrated in symbolic constraints and in one especially brittle model. A forked recall audit shows that 88.6% of observed failures are behavioral-drift failures rather than ordinary forgetting; excluding the brittle outlier preserves this diagnosis (84.4% of failures). Extended evaluations at depths up to 50 turns confirm that the dissociation persists beyond the core benchmark, while Structured State Anchoring recovers up to 24.4 points on immutable constraints. The results show that reliability cannot be inferred from single-turn instruction following alone; evaluation must track whether remembered constraints remain operational as dialogue state grows. Code and artifacts are available at: https://github.com/xzzvsxd/StateTrack.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Computer science, Suite, State (computer science), Computer security, Witness, Audit

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange