Metraipolis › archive › dc_f7308be55a86
TRACE: Trajectory Return Attribution and Contrastive Erasure for Multi-Turn Safety — arXiv:2610.01323
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2610.01323
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-06T12:04:38.603Z
- SHA-256 of the archived text
9a8f777fcf516d55cee83034f1cf61abe08f68e0aaeb62f3bf40f75671f2ef82- Stable URL of this page
https://metraipolis.com/documento/dc_f7308be55a86
How to cite
TRACE: Trajectory Return Attribution and Contrastive Erasure for Multi-Turn Safety — arXiv:2610.01323 — arXiv (en), https://arxiv.org/abs/2610.01323 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_f7308be55a86 (Metraipolis archive dc_f7308be55a86, sha256 9a8f777fcf516d55cee83034f1cf61abe08f68e0aaeb62f3bf40f75671f2ef82)
BibTeX
@misc{dcf7308be55a86,
title = {TRACE: Trajectory Return Attribution and Contrastive Erasure for Multi-Turn Safety — arXiv:2610.01323},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2610.01323},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_f7308be55a86; sha256 9a8f777fcf516d55cee83034f1cf61abe08f68e0aaeb62f3bf40f75671f2ef82},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.01323. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Safety-aligned large language models (LLMs) often refuse a harmful request but comply once the same goal is spread over several turns. Preference objectives score whole responses to single prompts, so their training loss alone cannot control risk on unseen histories. Our analysis gives sufficient conditions under which suppression at supervised single-turn contexts yields a bound on multi-turn trajectory risk. The bound accounts for coverage, transfer slack, and leakage, and characterizes contraction relative to a base-policy risk budget evaluated on the trained policy's contexts. TRACE (Trajectory Return Attribution and Contrastive Erasure) turns this principle into a token-level objective. On the safe response, each token is weighted by the discounted return of a refusal-attributable advantage. The advantage compares a frozen reference model with its refusal-ablated copy, allowing earlier response tokens to receive credit from later refusal-related evidence. At high-gap positions on rejected responses, TRACE combines the observed token with policy-selected alternatives in the erasure target. A gradient-norm penalty replaces the retain set. Across five open-weight models and seven multi-turn attacks, TRACE gives the lowest attack success rate (ASR) in all 35 model and attack pairs, while the model utility evaluated on MMLU and HellaSwag drop by at most 1\.23 points. Source code can be found in the supplemental material. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-10-02 in cs.AI). Categorie arXiv: Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2610.01323 — 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 2026-10-06T10:21:27.938Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-10-02?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).