Metraipolis › archive › dc_6526dff27670
SAVeR$^2$: Reasoning-based Safety Alignment for Large Reasoning Models via Verifiable Rewards — arXiv:2611.07482
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2611.07482
- License (as archived)
- CC0-1.0
- Archived
- 2026-11-06T14:41:36.760Z
- SHA-256 of the archived text
ad23af4b54ee0310d0390f6e9cfb33b290998e5b28d861a11b5c0995285534a8- Stable URL of this page
https://metraipolis.com/documento/dc_6526dff27670
How to cite
SAVeR$^2$: Reasoning-based Safety Alignment for Large Reasoning Models via Verifiable Rewards — arXiv:2611.07482 — arXiv (en), https://arxiv.org/abs/2611.07482 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_6526dff27670 (Metraipolis archive dc_6526dff27670, sha256 ad23af4b54ee0310d0390f6e9cfb33b290998e5b28d861a11b5c0995285534a8)
BibTeX
@misc{dc6526dff27670,
title = {SAVeR$^2$: Reasoning-based Safety Alignment for Large Reasoning Models via Verifiable Rewards — arXiv:2611.07482},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2611.07482},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_6526dff27670; sha256 ad23af4b54ee0310d0390f6e9cfb33b290998e5b28d861a11b5c0995285534a8},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.07482. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Large Reasoning Models (LRMs) have achieved impressive performance via explicit Chain-of-Thought (CoT) reasoning, yet this process introduces critical safety risks. We formalize LRM safety objectives as requiring non-malicious reasoning and answers, safety consistency, and explicit intent identification. We argue that existing alignment methods suffer from poor adversarial generalization, a significant ``safety tax'' on reasoning, and unfaithful coarse-grained rewards, primarily due to their neglect of the LRM's intrinsic reasoning capabilities for safety. To address these, we propose \textbf{SAVeR$^2$}, a novel \textbf{S}afety \textbf{A}lignment method via \textbf{R}ewards with \textbf{R}easoning capability. SAVeR$^2$ decomposes the safety objective into four fine-grained verifiable reward signals and utilizes a difficulty-aware data selection strategy to stabilize training. Crucially, we introduce a reasoning-preserving gradient projection mechanism that analytically resolves conflicts between safety and reasoning gradients, effectively eliminating the safety tax. Extensive experiments on models ranging from 1.5B to 671B demonstrate that SAVeR$^2$ achieves superior safety performance and significantly lower over-refusal rates across multiple benchmarks without compromising general reasoning capabilities.
Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2611.07482 — 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 2026-11-06T14:18:42.639Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).