mlsanityData to 7 Oct 2026

DeepSeek on a Trip: Inducing Targeted Visual Hallucinations via Representation Vulnerabilities

arXiv 2502.07905 ↗First posted 11 Feb 2025cs.CV cs.LGcited by 1

On the map: Multimodal LLMs / vision-language models → Hallucination mitigation in LVLMs

Abstract (arXiv)

Multimodal Large Language Models (MLLMs) represent the cutting edge of AI technology, with DeepSeek models emerging as a leading open-source alternative offering competitive performance to closed-source systems. While these models demonstrate remarkable capabilities, their vision-language integration mechanisms introduce specific vulnerabilities. We implement an adapted embedding manipulation attack on DeepSeek Janus that induces targeted visual hallucinations through systematic optimization of image embeddings. Through extensive experimentation across COCO, DALL-E 3, and SVIT datasets, we achieve hallucination rates of up to 98.0% while maintaining high visual fidelity (SSIM > 0.88) of the manipulated images on open-ended questions. Our analysis demonstrates that both 1B and 7B variants of DeepSeek Janus are susceptible to these attacks, with closed-form evaluation showing consistently higher hallucination rates compared to open-ended questioning. We introduce a novel multi-prompt hallucination detection framework using LLaMA-3.1 8B Instruct for robust evaluation. The implications of these findings are particularly concerning given DeepSeek's open-source nature and widespread deployment potential. This research emphasizes the critical need for embedding-level security measures in MLLM deployment pipelines and contributes to the broader discussion of responsible AI implementation.

Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Placed in Hallucination mitigation in LVLMs, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 29 of its 38 references that are in the map. First posted on arXiv 11 Feb 2025.

Builds on

29 of 38 references.

References · 29

  1. 2023GPT-4 Technical Reportcited by 18,132
  2. 2024The Llama 3 Herd of Modelscited by 14,192
  3. 2023Visual Instruction Tuningcited by 7,792
  4. 2014Explaining and Harnessing Adversarial Examplescited by 7,287
  5. 2017Towards Deep Learning Models Resistant to Adversarial Attackscited by 6,326
  6. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,146
  7. 2022Flamingo: a Visual Language Model for Few-Shot Learningcited by 4,337
  8. 2023Universal and Transferable Adversarial Attacks on Aligned Language Modelscited by 2,914
  9. 2023MiniGPT-4: Enhancing Vision-Language Understanding with Advanced Large Language Modelscited by 2,689
  10. 2015VQA: Visual Question Answeringcited by 2,489
  11. 2023PaLM-E: An Embodied Multimodal Language Modelcited by 2,037
  12. 2019Certified Adversarial Robustness via Randomized Smoothingcited by 1,324
17 more
  1. 2025Janus-Pro: Unified Multimodal Understanding and Generation with Data and Model Scalingcited by 787
  2. 2015The Limitations of Deep Learning in Adversarial Settingscited by 784
  3. 2016Universal adversarial perturbationscited by 774
  4. 2023OpenFlamingo: An Open-Source Framework for Training Large Autoregressive Vision-Language Modelscited by 560
  5. 2024Janus: Decoupling Visual Encoding for Unified Multimodal Understanding and Generationcited by 447
  6. 2018Audio Adversarial Examples: Targeted Attacks on Speech-to-Textcited by 268
  7. 2023Poisoning Language Models During Instruction Tuningcited by 254
  8. 2024JanusFlow: Harmonizing Autoregression and Rectified Flow for Unified Multimodal Understanding and Generationcited by 162
  9. 2023SVIT: Scaling up Visual Instruction Tuningcited by 161
  10. 2023On the Robustness of Large Multimodal Models Against Image Adversarial Attackscited by 89
  11. 2019Black-box Adversarial Attacks on Video Recognition Modelscited by 58
  12. 2024Test-Time Backdoor Attacks on Multimodal Large Language Modelscited by 35
  13. 2024Putting GPT-4o to the Sword: A Comprehensive Evaluation of Language, Vision, Speech, and Multimodal Proficiencycited by 31
  14. 2024Adversarial Attacks on Large Language Models Using Regularized Relaxationcited by 12
  15. 2024Malicious Path Manipulations via Exploitation of Representation Vulnerabilities of Vision-Language Navigation Systemscited by 10
  16. 2024Intriguing Equivalence Structures of the Embedding Space of Vision Transformerscited by 8
  17. 2025Mechanistic Understandings of Representation Vulnerabilities and Engineering Robust Vision Transformerscited by 2

Built on by

Cited by year

Citing papers per year: 2025: 1.

1 paper.

  1. 2025Innovating China's Intangible Cultural Heritage with DeepSeek + MidJourney: The Case of Yangliuqing theme Woodblock Printscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Exposing Hallucinations To Suppress Them: VLMs Representation Editing With Generative Anchors
  2. 2024Detecting and Mitigating Hallucination in Large Vision Language Models via Fine-Grained AI Feedback
  3. 2024Towards a Systematic Evaluation of Hallucinations in Large-Vision Language Models
  4. 2026HALP: Detecting Hallucinations in Vision-Language Models without Generating a Single Token
  5. 2023Detecting and Preventing Hallucinations in Large Vision Language Models
  6. 2026Beyond the Global Scores: Fine-Grained Token Grounding as a Robust Detector of LVLM Hallucinations
  7. 2024Hallucination in Multimodal Foundation Models: A Survey on Causes, Corrections, and Evaluations
  8. 2025Mirage in the Eyes: Hallucination Attack on Multi-modal Large Language Models with Only Attention Sink