Metraipolis › archive › dc_08896e0f4fcf

Safety Does Not Compose: Non-Decaying Loop State for Autonomous LLM Agents — arXiv:2608.27141

Source
arXiv (en)
Original
https://arxiv.org/abs/2608.27141
License (as archived)
CC0-1.0
Archived
2026-10-17T14:30:35.337Z
SHA-256 of the archived text
4dc7017dfd120d3f288783c778c1bceef8e8cd1074f59cca6c33621575a3fed9
Stable URL of this page
https://metraipolis.com/documento/dc_08896e0f4fcf

How to cite

Safety Does Not Compose: Non-Decaying Loop State for Autonomous LLM Agents — arXiv:2608.27141 — arXiv (en), https://arxiv.org/abs/2608.27141 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_08896e0f4fcf (Metraipolis archive dc_08896e0f4fcf, sha256 4dc7017dfd120d3f288783c778c1bceef8e8cd1074f59cca6c33621575a3fed9)
BibTeX
@misc{dc08896e0f4fcf,
  title = {Safety Does Not Compose: Non-Decaying Loop State for Autonomous LLM Agents — arXiv:2608.27141},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2608.27141},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_08896e0f4fcf; sha256 4dc7017dfd120d3f288783c778c1bceef8e8cd1074f59cca6c33621575a3fed9},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.27141. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Large language model agents are increasingly deployed as autonomous loops. Starting from one human goal, such a system repeatedly discovers work, plans, executes tool calls, verifies outcomes and persists state across many unattended iterations. The agent safeguards in wide use, however, are defined over a single trajectory, and their safety state is re-initialized when the next trajectory begins. We show that this is a failure of composition rather than an implementation detail. Our central result is a separation: against an attack whose evidence is fragmented across several iterations, every trajectory-scoped monitor has a true-positive rate equal to its false-positive rate, however expressive it is, because the evidence it would need never appears in the window it sees, whereas a monitor retaining cross-iteration state separates the two perfectly. We further show that the obvious repair of carrying a geometrically decaying risk score is insufficient, because the cooling-off period a patient adversary must wait is a constant that does not grow with the horizon $N$. We then present LoopHarness, which restores a persistent, non-decaying safety state at the loop level. Under mediated commits and an arbiter detection floor $\delta_M$, it bounds the expected number of unauthorized irreversible actions by $B+m-1+m/\delta_M$, a constant in $N$, of which the $B+m-1$ term is decided by a model-free rule and therefore survives a fully colluding verifier. We give a complete evaluation protocol on native Agent-SafetyBench tasks with paired clean and attacked episodes, an outer-state attack suite whose decisive evidence exists only across iterations, per-module ablations, and an adaptive white-box red team.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-28 in cs.AI).
Categorie arXiv: Cryptography and Security (cs.CR) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2608.27141 — 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 2026-10-17T13:16:53.693Z 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).