mlsanityData to 7 Oct 2026

Typographic Attack Against VLM-based AI-generated Image Detection

arXiv 2609.39662 ↗First posted 30 Sep 2026cs.CVcited by 0

On the map: Multimodal LLMs / vision-language models → MLLM jailbreaks and adversarial robustness

Abstract (arXiv)

Vision-language models (VLMs) are increasingly used for AI-generated image (AIGI) detection, providing natural-language explanations for authenticity judgments. However, their ability to interpret text within images may also expose these judgments to misleading semantic cues. We systematically evaluate typographic attack strategies across detection-oriented, open-weight, and commercial VLMs, considering both real-to-fake and fake-to-real attacks. Our results show that reasoning modes generally exhibit greater vulnerability than direct modes and that attack effectiveness exhibits pronounced directional asymmetry. Moreover, larger models tend to exhibit higher clean detection accuracy but also higher attack success rates. We further examine attack robustness under image and text transformations and investigate whether overlays indicating the correct class can aid error correction. Together, these analyses characterize how typographic attacks influence authenticity judgments and expose limitations of current VLM-based AIGI detection systems.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in MLLM jailbreaks and adversarial robustness, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 24 of its 28 references that are in the map. First posted on arXiv 30 Sep 2026.

Builds on

24 of 28 references.

Closest ancestors · 4

Same direction, newest first

  1. 2025Spot the Fake: Large Multimodal Model-Based Synthetic Image Detection with Artifact Explanation19 Mar
  2. 2025Web Artifact Attacks Disrupt Vision Language Models17 Mar
  3. 2024Vision-LLMs Can Fool Themselves with Self-Generated Typographic Attacks1 Feb
  4. 2023FigStep: Jailbreaking Large Vision-Language Models via Typographic Visual Prompts9 Nov

Other references · 20

  1. 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,767
  2. 2023PaLM-E: An Embodied Multimodal Language Modelcited by 2,037
  3. 2019Celeb-DF: A Large-scale Challenging Dataset for DeepFake Forensicscited by 573
  4. 2025GLM-4.5V and GLM-4.1V-Thinking: Towards Versatile Multimodal Reasoning with Scalable Reinforcement Learningcited by 439
  5. 2023GenImage: A Million-Scale Benchmark for Detecting AI-Generated Imagecited by 318
  6. 2017Synthetic and Natural Noise Both Break Neural Machine Translationcited by 304
  7. 2023Multilingual Jailbreak Challenges in Large Language Modelscited by 222
  8. 2024A Sanity Check for AI-generated Image Detectioncited by 187
  9. 2026Qwen3.5-Omni Technical Reportcited by 146
  10. 2024ScreenAI: A Vision-Language Model for UI and Infographics Understandingcited by 97
  11. 2025Words or Vision: Do Vision-Language Models Have Blind Faith in Text?cited by 86
  12. 2023VLATTACK: Multimodal Adversarial Attacks on Vision-Language Tasks via Pre-trained Modelscited by 66
8 more
  1. 2025Ivy-Fake: A Unified Explainable Framework and Benchmark for Image and Video AIGC Detectioncited by 34
  2. 2025BusterX++: Towards Unified Cross-Modal AI-Generated Content Detection and Explanation with MLLMcited by 28
  3. 2025Veritas: Generalizable Deepfake Detection via Pattern-Aware Reasoningcited by 23
  4. 2021Amicable Aid: Perturbing Images to Improve Classification Performancecited by 3
  5. 2024Nearly Solved? Robust Deepfake Detection Requires More than Visual Forensicscited by 2
  6. 2025Beyond Pixels: Semantic-aware Typographic Attack for Geo-Privacy Protectioncited by 2
  7. 2026Veritas++: Value-aware On-Policy Distillation for Perception-Enhanced AIGI Detectioncited by 2
  8. 2026DRIFT: Derailing Denoising Trajectories of Flow-Matching VLAs with Adversarial Patch Attackcited by 1

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2024Unveiling Typographic Deceptions: Insights of the Typographic Vulnerability in Large Vision-Language Model
  2. 2025SCAM: A Real-World Typographic Robustness Evaluation for Multimodal Foundation Models
  3. 2026Reading Between the Pixels: Linking Text-Image Embedding Alignment to Typographic Attack Success on Vision-Language Models
  4. 2024Harnessing the Power of Large Vision Language Models for Synthetic Image Detection
  5. 2026QuISE: Defense against Typographic Attacks on VLMs via Query-Irrelevant Semantic Editing
  6. 2026One Perturbation, Two Failure Modes: Probing VLM Safety via Embedding-Guided Typographic Perturbations
  7. 2025Towards Explainable Fake Image Detection with Multi-Modal Large Language Models
  8. 2024TrojVLM: Backdoor Attack Against Vision Language Models