mlsanityData to 14 Oct 2026

The Side Effects of Being Smart: Safety Risks in MLLMs' Multi-Image Reasoning

arXiv 2601.14127 ↗First posted 20 Jan 2026cs.CV cs.CLcited by 3

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

also in 1 more
Abstract (arXiv)

As Multimodal Large Language Models (MLLMs) acquire stronger reasoning capabilities to handle complex, multi-image instructions, this advancement may pose new safety risks. We study this problem by introducing MIR-SafetyBench, the first benchmark focused on multi-image reasoning safety, which consists of 2,676 instances across a taxonomy of 9 multi-image relations. Our extensive evaluations on 19 MLLMs reveal a troubling trend: models with more advanced multi-image reasoning can be more vulnerable on MIR-SafetyBench. Beyond attack success rates, we find that many responses labeled as safe are superficial, often driven by misunderstanding or evasive, non-committal replies. We further observe that unsafe generations exhibit lower attention entropy than safe ones on average. This internal signature suggests a possible risk that models may over-focus on task solving while neglecting safety constraints. Our code and data are available at https://github.com/thu-coai/MIR-SafetyBench.

Key facts. Cited by 3 papers in the map (data through 14 Oct 2026). Placed in Reasoning for agents, planning and code, one of 17 directions in LLM reasoning. Builds on 30 of its 58 references that are in the map. First posted on arXiv 20 Jan 2026.

Builds on

30 of 58 references.

Closest ancestors · 3

Same direction, newest first

  1. 2025Chain-of-Thought Hijacking30 Oct
  2. 2025VisCRA: A Visual Chain Reasoning Attack for Jailbreaking Multimodal Large Language Models26 May
  3. 2025Rethinking Bottlenecks in Safety Fine-Tuning of Vision Language Models30 Jan

Other references · 27

  1. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,330
  2. 2024GPT-4o System Cardcited by 4,588
  3. 2025Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilitiescited by 3,796
  4. 2023Universal and Transferable Adversarial Attacks on Aligned Language Modelscited by 3,104
  5. 2025InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Modelscited by 1,834
  6. 2024HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusalcited by 1,429
  7. 2024MiniCPM-V: A GPT-4V Level MLLM on Your Phonecited by 1,079
  8. 2025FLUX.1 Kontext: Flow Matching for In-Context Image Generation and Editing in Latent Spacecited by 1,006
  9. 2025Vision-R1: Incentivizing Reasoning Capability in Multimodal Large Language Modelscited by 665
  10. 2025GLM-4.5V and GLM-4.1V-Thinking: Towards Versatile Multimodal Reasoning with Scalable Reinforcement Learningcited by 466
  11. 2025Kimi-VL Technical Reportcited by 388
  12. 2024JailBreakV: A Benchmark for Assessing the Robustness of MultiModal Large Language Models against Jailbreak Attackscited by 230
15 more
  1. 2025Grounded Chain-of-Thought for Multimodal Large Language Modelscited by 59
  2. 2024Multimodal Situational Safetycited by 58
  3. 2024Visual Haystacks: A Vision-Centric Needle-In-A-Haystack Benchmarkcited by 55
  4. 2024MMAD: A Comprehensive Benchmark for Multimodal Large Language Models in Industrial Anomaly Detectioncited by 47
  5. 2024Heuristic-Induced Multimodal Risk Distribution Jailbreak Attack for Multimodal Large Language Modelscited by 33
  6. 2025What Prompts Don't Say: Understanding and Managing Underspecification in LLM Promptscited by 32
  7. 2025Skywork-R1V3 Technical Reportcited by 16
  8. 2024Cognitive Overload Attack:Prompt Injection for Long Contextcited by 13
  9. 2025AISafetyLab: A Comprehensive Framework for AI Safety Evaluation and Improvementcited by 12
  10. 2025LongSafety: Evaluating Long-Context Safety of Large Language Modelscited by 12
  11. 2025MAC: A Live Benchmark for Multimodal Large Language Models in Scientific Understandingcited by 12
  12. 2025United Minds or Isolated Agents? Exploring Coordination of LLMs under Cognitive Load Theorycited by 10
  13. 2025ShieldVLM: Safeguarding the Multimodal Implicit Toxicity via Deliberative Reasoning with LVLMscited by 7
  14. 2024Self-Attention Limits Working Memory Capacity of Transformer-Based Modelscited by 5
  15. 2025ELITE: Enhanced Language-Image Toxicity Evaluation for Safetycited by 5

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 3.

3 papers.

  1. 2026Safe Alone, Unsafe Together: Safeguarding Against Implicit Toxicity When Benign Images Combinecited by 0
  2. 2026Automatic Hard Example Synthesis with Multi-Level Agentic Data Curationcited by 0
  3. 2026Scenes as an Attack Surface: Uncovering Scene-Grounded Safety Failures in MLLMscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025SafeMLRM: Demystifying Safety in Multi-modal Large Reasoning Models
  2. 2026When Helpers Become Hazards: A Benchmark for Analyzing Multimodal LLM-Powered Safety in Daily Life
  3. 2025VLSU: Mapping the Limits of Joint Multimodal Understanding for AI Safety
  4. 2025When Safe Unimodal Inputs Collide: Optimizing Reasoning Chains for Cross-Modal Safety in Multimodal Large Language Models
  5. 2026MME-Safety: A Fine-grained Benchmark for Safety Evaluation of MLLMs
  6. 2025MSTS: A Multimodal Safety Test Suite for Vision-Language Models
  7. 2025Can't See the Forest for the Trees: Benchmarking Multimodal Safety Awareness for Multimodal LLMs
  8. 2025SafeMT: Multi-turn Safety for Multimodal Language Models