Sorry Robot, Happy Human: Vision-Language Models Read Only One of Two Legible Typographic Layers

arXiv:2609.31403 · cs.CL, cs.AI, cs.CV · Submitted 2026-09-25 · Read on arXiv

cs.CL, cs.AI, cs.CV

Submitted: 2026-09-25

Updated: 2026-09-25

Code: https://github.com/yesdopepe/DecoyBench

Terminology

• Accuracy — 13× in this paper · explained in 1 episode
• Verification — 3× in this paper · explained in 2 episodes
• Benchmark — 2× in this paper · explained in 2 episodes
• Robustness — 2× in this paper · explained in 5 episodes
• Vision-Language Models (VLMs) — 2× in this paper · explained in 6 episodes
• Commentary — 1× in this paper · explained in 2 episodes
• Exact Match — 1× in this paper
• Foundation Model — 1× in this paper · explained in 3 episodes
• Foundation Models — 1× in this paper · explained in 7 episodes
• Generalizability — 1× in this paper · explained in 6 episodes
• Levenshtein Similarity — 1× in this paper

Sources

• ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness
• Invisible Injections: Exploiting Vision-Language Models Through Steganographic Prompt Embedding
• Vision-LLMs Can Fool Themselves with Self-Generated Typographic Attacks
• SCAM: A Real-World Typographic Robustness Evaluation for Multimodal Foundation Models

Related papers

• Exploring Solution Divergence and Its Effect on Large Language Model Problem Solving
• Ishigaki-IDS-Bench: A Benchmark for Generating Information Delivery Specification from BIM Information Requirements
• Subliminal Steering: Stronger Encoding of Hidden Signals
• MedStruct-S: A Benchmark for Key Discovery, Key-Conditioned QA and Semi-Structured Extraction from OCR Clinical Reports
• The End of Transformers? On Challenging Attention and the Rise of Sub-Quadratic Architectures
• Untangling the Mechanisms of Misleading Context in Medical Question Answering