List your company

Prompt Injection Attack on PaperGuard

80Attack Success Rate (ASR)Claude-sonnet-4.5
44.6453.826372.18Jun 10, 2026

Evaluation results

MethodLinks
Claude-sonnet-4.5Multimodality=true2026.068014.141.13-2.62
Mistral-Small-3.1-24B-InstructMultimodality=true2026.06769.950-3.26
Gemma-3-27b-itMultimodality=true2026.06747.930.27-1.88
Qwen2.5-VL-32B-InstructMultimodality=true2026.06737.110-2.43
GPT-4oMultimodality=true2026.06726.650.13-1.37
GPT-4.1Multimodality=false2026.06689.40.43-2.27
Qwen-3-8BMultimodality=false2026.06644.160.3-1.39
DeepSeek-R1-Distill-Llama-8BMultimodality=false2026.06465.050.460.11