Metraipolis › archive › dc_27ffecfb4715

Approved Too Late: Verdict Staleness in LLM-Guarded Self-Adaptive Systems — arXiv:2608.26306

Source
arXiv (en)
Original
https://arxiv.org/abs/2608.26306
License (as archived)
CC0-1.0
Archived
2026-10-17T14:22:32.610Z
SHA-256 of the archived text
11b1c87db16c91669a75fc3fc147a485750b9adb9d2442b33d1528327a197179
Stable URL of this page
https://metraipolis.com/documento/dc_27ffecfb4715

How to cite

Approved Too Late: Verdict Staleness in LLM-Guarded Self-Adaptive Systems — arXiv:2608.26306 — arXiv (en), https://arxiv.org/abs/2608.26306 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_27ffecfb4715 (Metraipolis archive dc_27ffecfb4715, sha256 11b1c87db16c91669a75fc3fc147a485750b9adb9d2442b33d1528327a197179)
BibTeX
@misc{dc27ffecfb4715,
  title = {Approved Too Late: Verdict Staleness in LLM-Guarded Self-Adaptive Systems — arXiv:2608.26306},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2608.26306},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_27ffecfb4715; sha256 11b1c87db16c91669a75fc3fc147a485750b9adb9d2442b33d1528327a197179},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.26306. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

A large language model (LLM) guardrail for a self-adaptive system (SAS) may issue an approval that is correct at check time but stale by actuation. This creates an Execute-stage time-of-check to time-of-use (TOCTOU) hazard. We study verdict freshness: whether a guardrail verdict remains valid when used. We distinguish three quantities that answer different questions: all-candidate verdict change under fixed-action replay, oracle-labeled approval expiry on recorded closed-loop trajectories, and judge-conditioned use-time invalidity. Across five reproducible SAS environments, all-candidate verdict-change rates span 5.3-48.4% at a common replay shift of eight simulator steps. We introduce the Freshness-Bounded Shield (FBS), which estimates each approval's validity horizon from its safe-side margin and recent feature volatility, without an explicit plant-dynamics model. Using fixed settings documented in the artifact, FBS reduces oracle-labeled approval-expiry rates from 3.4-24.7% to 0-1.8% at the same shift. A separate audit of four LLM judges finds nonzero judge-conditioned use-time invalidity in every approval stream. We formulate a freshness contract: every approval must be correct at check time and remain valid at use time.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-28 in cs.AI).
Categorie arXiv: Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2608.26306 — 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 2026-10-17T13:16:53.485Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-08-28?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).