mlsanityData to 7 Oct 2026

SafeMLRM: Demystifying Safety in Multi-modal Large Reasoning Models

arXiv 2504.08813 ↗First posted 9 Apr 2025cs.LG cs.AI cs.CRcited by 29

On the map: LLM reasoning → Reasoning for agents, planning and code

also in 2 more
Abstract (arXiv)

The rapid advancement of multi-modal large reasoning models (MLRMs) -- enhanced versions of multimodal 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 times higher attack rates -- far exceeding the average 3.4 times 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.

Key facts. Cited by 29 papers in the map (data through 7 Oct 2026). Placed in Reasoning for agents, planning and code, one of 17 directions in LLM reasoning. Builds on 29 of its 55 references that are in the map. First posted on arXiv 9 Apr 2025.

Builds on

29 of 55 references.

Closest ancestors · 6

Same direction, newest first

  1. 2025Safety Tax: Safety Alignment Makes Your Large Reasoning Models Less Reasonable1 Mar
  2. 2025The Hidden Risks of Large Reasoning Models: A Safety Assessment of R118 Feb
  3. 2025Safety Evaluation of DeepSeek Models in Chinese Contexts16 Feb
  4. 2025Safety Reasoning with Guidelines6 Feb
  5. 2025The dark deep side of DeepSeek: Fine-tuning attacks against the safety alignment of CoT-enabled models3 Feb
  6. 2025o3-mini vs DeepSeek-R1: Which One is Safer?30 Jan

Other references · 23

  1. 2023GPT-4 Technical Reportcited by 18,132
  2. 2023Llama 2: Open Foundation and Fine-Tuned Chat Modelscited by 12,865
  3. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 8,851
  4. 2024GPT-4o System Cardcited by 4,467
  5. 2024Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolutioncited by 4,151
  6. 2024HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusalcited by 1,330
  7. 2024DeepSeek-Coder: When the Large Language Model Meets Programming -- The Rise of Code Intelligencecited by 1,154
  8. 2023FigStep: Jailbreaking Large Vision-Language Models via Typographic Visual Promptscited by 368
  9. 2025R1-Onevision: Advancing Generalized Multimodal Reasoning through Cross-Modal Formalizationcited by 364
  10. 2025Multimodal Chain-of-Thought Reasoning: A Comprehensive Surveycited by 176
  11. 2024Mulberry: Empowering MLLM with o1-like Reasoning and Reflection via Collective Monte Carlo Tree Searchcited by 171
  12. 2024Improve Vision Language Model Chain-of-thought Reasoningcited by 135
11 more
  1. 2024Insight-V: Exploring Long-Chain Visual Reasoning with Multimodal Large Language Modelscited by 112
  2. 2024Jailbreak Large Vision-Language Models Through Multi-Modal Linkagecited by 53
  3. 2025VisRL: Intention-Driven Visual Perception via Reinforced Reasoningcited by 39
  4. 2024TextCoT: Zoom In for Enhanced Multimodal Text-Rich Image Understandingcited by 37
  5. 2024Thinking Before Looking: Improving Multimodal LLM Reasoning via Mitigating Visual Hallucinationcited by 29
  6. 2025Towards Understanding the Safety Boundaries of DeepSeek Models: Evaluation and Findingscited by 24
  7. 2023Chain of Images for Intuitively Reasoningcited by 20
  8. 2024PKRD-CoT: A Unified Chain-of-thought Prompting for Multi-Modal Large Language Models in Autonomous Drivingcited by 17
  9. 2025Challenges in Ensuring AI Safety in DeepSeek-R1 Models: The Shortcomings of Reinforcement Learning Strategiescited by 11
  10. 2025Leveraging Chain of Thought towards Empathetic Spoken Dialogue without Corresponding Question-Answering Datacited by 7
  11. 2023Let's Think Frame by Frame with VIP: A Video Infilling and Prediction Dataset for Evaluating Video Chain-of-Thoughtcited by 5

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 25; 2026: 4.

29 papers.

Sort:

By area:

  1. 2025Safety in Large Reasoning Models: A Surveycited by 58
  2. 2025LinearRAG: Linear Graph Retrieval Augmented Generation on Large-scale Corporacited by 47
  3. 2025Don't Overthink It: A Survey of Efficient R1-style Large Reasoning Modelscited by 36
  4. 2025AGENTSAFE: Benchmarking the Safety of Embodied Agents on Hazardous Instructionscited by 33
  5. 2025SAFEPATH: Preventing Harmful Reasoning in Chain-of-Thought via Early Alignmentcited by 24
  6. 2025You Don't Need Pre-built Graphs for RAG: Retrieval Augmented Generation with Adaptive Reasoning Structurescited by 23
  7. 2025A Comprehensive Survey on Trustworthiness in Reasoning with Large Language Modelscited by 22
  8. 2025DocPruner: A Storage-Efficient Framework for Multi-Vector Visual Document Retrieval via Adaptive Patch-Level Embedding Pruningcited by 19
  9. 2025Think in Safety: Unveiling and Mitigating Safety Alignment Collapse in Multimodal Large Reasoning Modelcited by 18
  10. 2025RSafe: Incentivizing proactive reasoning to build robust and adaptive LLM safeguardscited by 14
  11. 2025SafeMobile: Chain-level Jailbreak Detection and Automated Evaluation for Multimodal Mobile Agentscited by 14
  12. 2025Enhancing CLIP Robustness via Cross-Modality Alignmentcited by 14
17 more
  1. 2025Pushing the Limits of Safety: A Technical Report on the ATLAS Challenge 2025cited by 11
  2. 2025The Eye of Sherlock Holmes: Uncovering User Private Attribute Profiling via Vision-Language Model Agentic Frameworkcited by 9
  3. 2025MLLMEraser: Achieving Test-Time Unlearning in Multimodal Large Language Models through Activation Steeringcited by 9
  4. 2025Adaptive Multi-prompt Contrastive Network for Few-shot Out-of-distribution Detectioncited by 8
  5. 2025SaFeR-VLM: Toward Safety-aware Fine-grained Reasoning in Multimodal Modelscited by 8
  6. 2025Path Drift in Large Reasoning Models:How First-Person Commitments Override Safetycited by 5
  7. 2025GuardTrace-VL: Detecting Unsafe Multimodel Reasoning via Iterative Safety Supervisioncited by 4
  8. 2026Reasoning over Precedents Alongside Statutes: Case-Augmented Deliberative Alignment for LLM Safetycited by 3
  9. 2026Mitigating Safety Tax via Distribution-Grounded Refinement in Large Reasoning Modelscited by 3
  10. 2026Safety Recovery in Reasoning Models Is Only a Few Early Steering Steps Awaycited by 3
  11. 2025Red Teaming Large Reasoning Modelscited by 2
  12. 2025Beyond SFT: Reinforcement Learning for Safer Large Reasoning Models with Better Reasoning Abilitycited by 2
  13. 2025PD3F: A Pluggable and Dynamic DoS-Defense Framework Against Resource Consumption Attacks Targeting Large Language Modelscited by 1
  14. 2025From Perception to Reasoning: Deep Thinking Empowers Multimodal Large Language Modelscited by 1
  15. 2025COSMO-RL: Towards Trustworthy LMRMs via Joint Safety and Stabilitycited by 0
  16. 2025VisuoAlign: Safety Alignment of LVLMs with Multimodal Tree Searchcited by 0
  17. 2026Feature-Space Smoothing: Certified Robustness of Deep Representationscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025When Safe Unimodal Inputs Collide: Optimizing Reasoning Chains for Cross-Modal Safety in Multimodal Large Language Models
  2. 2026Evolving Safety Landscape of Multi-modal Large Language Models: A Survey of Emerging Threats and Safeguards
  3. 2026MME-Safety: A Fine-grained Benchmark for Safety Evaluation of MLLMs
  4. 2026The Side Effects of Being Smart: Safety Risks in MLLMs' Multi-Image Reasoning
  5. 2026On the Efficiency-Safety Dilemma in Large Reasoning Models
  6. 2025SafeRBench: Dissecting the Reasoning Safety of Large Language Models
  7. 2026Beyond Content Safety: Real-Time Monitoring for Reasoning Vulnerabilities in Large Language Models
  8. 2025USB: A Comprehensive and Unified Safety Evaluation Benchmark for Multimodal Large Language Models