EUVIMEDEUVIMED entdecken
Crossref

SafeMLRM: Demystifying safety in multi-modal large reasoning models

Junfeng Fang, Yukai Wang, Ruipeng Wang, Xiang Wang, Xiangnan He

AI Plus · 2026

DOI: 10.59717/ipj.aiplus.2026.100014 ↗

Kurz gefasst

Fachartikel in der Zeitschrift AI Plus, 2026. Verfasst von Junfeng Fang und 4 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 53 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The rapid advancement of multi-modal large reasoning models (MLRMs) — enhanced versions of multi-modal large language models (MLLMs) equipped with reasoning capabilities — has revolutionized diverse applications. However, their safety implications remain underexplored. While prior work has exposed critical vulnerabilities in unimodal reasoning models, MLRMs introduce distinct risks from cross-modal reasoning pathways. This work presents the first systematic safety analysis of MLRMs through large-scale empirical studies comparing MLRMs with their base MLLMs. Our experiments reveal three critical findings: (1) The Reasoning Tax: Acquiring reasoning capabilities catastrophically degrades inherited safety alignment. MLRMs exhibit 37.44% higher jailbreaking success rates than base MLLMs under adversarial attacks. (2) Safety Blind Spots: While safety degradation is pervasive, certain scenarios (e.g., Illegal Activity) suffer 25×higher attack rates — far exceeding the average 3.4×increase, revealing scenario-specific vulnerabilities with alarming cross-model and datasets consistency. (3) Emergent Self-Correction: Despite tight reasoning-answer safety coupling, MLRMs demonstrate nascent self-correction — 16.9% of jailbroken reasoning steps are overridden by safe answers, hinting at intrinsic safeguards. These findings underscore the urgency of scenario-aware safety auditing and mechanisms to amplify MLRMs’ self-correction potential. To catalyze research, we open-source OpenSafeMLRM, the first toolkit for MLRM safety evaluation, providing unified interface for mainstream models, datasets, and jailbreaking methods. Our work calls for immediate efforts to harden reasoning-augmented AI, ensuring its transformative potential aligns with ethical safeguards.

Zitieren

APA

Fang, J., Wang, Y., Wang, R., Wang, X., & He, X. (2026). SafeMLRM: Demystifying safety in multi-modal large reasoning models. AI Plus. https://doi.org/10.59717/ipj.aiplus.2026.100014

Vancouver

Fang J, Wang Y, Wang R, Wang X, He X. SafeMLRM: Demystifying safety in multi-modal large reasoning models. AI Plus. 2026. doi:10.59717/ipj.aiplus.2026.100014

BibTeX
@article{Fang2026,
  author = {Fang, Junfeng and Wang, Yukai and Wang, Ruipeng and Wang, Xiang and He, Xiangnan},
  title = {SafeMLRM: Demystifying safety in multi-modal large reasoning models},
  journal = {AI Plus},
  year = {2026},
  doi = {10.59717/ipj.aiplus.2026.100014}
}
Kostenlos starten →