List your company

Jailbreak Attack on HADES

99.6Attack Success RateDMN
3.597628.521353.44578.3687Feb 28, 2026Mar 20, 2026Apr 10, 2026May 1, 2026May 22, 2026Jun 12, 2026Jul 3, 2026

Evaluation results

MethodLinks
DMNTarget Model=GPT-4V2026.0599.6————
DMNTarget Model=GPT-4o2026.0599.47————
DMNTarget Model=Seed1.5-VL2026.0599.2————
DMNTarget Model=Gemini-2.5-pro2026.0598.67————
INFER+Target Model=Qwen2-VL, Model Type=Open-source2026.0797.52————
MIDASTarget Model=Qwen2.5-VL2026.0297.36———3.65
INFER+Target Model=InternVL2, Model Type=Open-source2026.0796.21————
DMNTarget Model=Gemini-2.5-flash2026.0595.87————
DMNTarget Model=Qwen3-VL-plus2026.0595.73————
DMNTarget Model=Claude Sonnet 42026.0594.53————
INFER+Target Model=Gemini 2.5 Flash Lite, Model Type=Closed-source, Surrogate Model=Qwen2-VL2026.0794.31————
MIDASTarget Model=QVQ-Max2026.0294.24———4.21
INFERTarget Model=Qwen2-VL, Model Type=Open-source2026.0794.02————
MIDASTarget Model=Gemini-2.5-FT2026.0293.34———4.32
DMNTarget Model=Average2026.0593.09————
INFER+Target Model=InternVL3.5, Model Type=Open-source2026.0792.86————
INFER+Target Model=Llama 3.2-Vision, Model Type=Open-source2026.0792.57————
INFER+Target Model=GPT-4.1 mini, Model Type=Closed-source, Surrogate Model=Qwen2-VL2026.0791.11————
DMNTarget Model=o4-mini2026.0588.4————
INFERTarget Model=InternVL2, Model Type=Open-source2026.0787.61————
CS-DJTarget Model=Qwen2-VL, Model Type=Open-source2026.0786.59————
INFERTarget Model=Llama 3.2-Vision, Model Type=Open-source2026.0786.15————
CS-DJTarget Model=InternVL2, Model Type=Open-source2026.0785.57————
MIDASTarget Model=Gemini-2.5-Pro2026.0284.55———3.66
INFER+Target Model=Qwen3-VL, Model Type=Open-source2026.0784.55————
INFERTarget Model=InternVL3.5, Model Type=Open-source2026.0784.11————
TVPITarget Model=Qwen2-VL, Model Type=Open-source2026.0783.82————
CS-DJTarget Model=InternVL3.5, Model Type=Open-source2026.0782.94————
DMNTarget Model=GPT-52026.0582————
INFERTarget Model=Gemini 2.5 Flash Lite, Model Type=Closed-source2026.0781.63————
INFER+Target Model=Gemini 2.5 Flash, Model Type=Closed-source, Surrogate Model=Qwen2-VL2026.0780.32————
DMNTarget Model=o32026.0577.47————
LLaVA-1.5-7B (Baseline)VLM=LLaVA-1.5-7B2026.0373.7————
MIDASTarget Model=GPT-5-Chat2026.0272.18———3.12
CMRMVLM=LLaVA-1.5-7B2026.0371.4————
INFERTarget Model=Qwen3-VL, Model Type=Open-source2026.0769.97————
VisCRATarget Model=QVQ-Max2026.0265.87———2.83
HIMRDTarget Model=Qwen2.5-VL2026.0265.8———2.47
ShiftDCVLM=LLaVA-1.5-7B2026.0365.3————
CS-DJTarget Model=GPT-4.1 mini, Model Type=Closed-source2026.0765.01————
ImageTypoTarget Model=Qwen2-VL, Model Type=Open-source2026.0764.69————
INFERTarget Model=GPT-4.1 mini, Model Type=Closed-source2026.0761.66————
MIDASTarget Model=GPT-4o2026.0261.49———2.48
MIDASTarget Model=InternVL2.52026.0259.44———2.1
CS-DJTarget Model=Llama 3.2-Vision, Model Type=Open-source2026.0758.6————
CS-DJTarget Model=Gemini 2.5 Flash Lite, Model Type=Closed-source2026.0755.69————
ShareGPT4V-7B (Baseline)VLM=ShareGPT4V-7B2026.0355.1————
INFERTarget Model=Gemini 2.5 Flash, Model Type=Closed-source2026.0752.92————
TVPITarget Model=Qwen3-VL, Model Type=Open-source2026.0751.75————
CS-DJTarget Model=Gemini 2.5 Flash, Model Type=Closed-source2026.0750.87————
VisCRATarget Model=Gemini-2.5-FT2026.0246.98———2.22
TVPITarget Model=InternVL2, Model Type=Open-source2026.0743.73————
HADESTarget Model=QVQ-Max2026.0243.37———1.74
VideoJailTarget Model=Qwen3-VL-plus2026.0541.73————
ImageTypoTarget Model=InternVL2, Model Type=Open-source2026.0741.08————
HIMRDTarget Model=Gemini-2.5-Pro2026.0240.2———1.43
HIMRDTarget Model=GPT-4o2026.0240.1———1.46
FigStepTarget Model=QVQ-Max2026.0238.78———1.61
VisCRATarget Model=GPT-4o2026.0234.28———1.34
ImageTypoTarget Model=InternVL3.5, Model Type=Open-source2026.0729.74————
ECSOVLM=ShareGPT4V-7B2026.0329.6————
SI-AttackTarget Model=Qwen2.5-VL2026.0229.32———1.31
CS-DJTarget Model=Qwen3-VL, Model Type=Open-source2026.0729.01————
ECSOVLM=LLaVA-1.5-7B2026.0327.6————
AdaShieldVLM=LLaVA-1.5-7B2026.0326.5————
ShiftDCVLM=InternVL-Chat-19B2026.0325.2————
InternVL-Chat-19B (Baseline)VLM=InternVL-Chat-19B2026.0324.9————
SI-AttackTarget Model=InternVL2.52026.0224.79———1.16
ShiftDCVLM=ShareGPT4V-7B2026.0322.2————
VideoJailTarget Model=Seed1.5-VL2026.0522————
VisCRATarget Model=InternVL2.52026.0219.68———0.66
TVPITarget Model=Llama 3.2-Vision, Model Type=Open-source2026.0718.66————
TVPITarget Model=Gemini 2.5 Flash, Model Type=Closed-source2026.0718.51————
CMRMVLM=InternVL-Chat-19B2026.0317.6————
TVPITarget Model=InternVL3.5, Model Type=Open-source2026.0717.35————
HIMRDTarget Model=QVQ-Max2026.0217———0.93
ECSOVLM=InternVL-Chat-19B2026.0316.8————
HADES (method)Target Model=Seed1.5-VL2026.0516.8————
ImageTypoTarget Model=Gemini 2.5 Flash, Model Type=Closed-source2026.0715.89————
FigStepTarget Model=InternVL2.52026.0214.86———0.48
HADESTarget Model=InternVL2.52026.0214.86———0.48
JRS-RemVLM=InternVL-Chat-19B2026.0313.6————
AdaShieldVLM=InternVL-Chat-19B2026.0313.4————
HADESTarget Model=Gemini-2.5-Pro2026.0212.44———0.53
AdaShieldVLM=ShareGPT4V-7B2026.0312.1————
VisCRATarget Model=Gemini-2.5-Pro2026.0212.05———0.74
SI-AttackTarget Model=QVQ-Max2026.0211.89———1.01
HADESTarget Model=Qwen2.5-VL2026.0210.04———0.42
VisCRATarget Model=Qwen2.5-VL2026.0210.04———0.43
JRS-RemVLM=LLaVA-1.5-7B2026.039.4————
ImageTypoTarget Model=Llama 3.2-Vision, Model Type=Open-source2026.079.33————
HIMRDTarget Model=GPT-5-Chat2026.028.8———0.37
HADES (method)Target Model=Gemini-2.5-flash2026.058.8————
VideoJailTarget Model=Average2026.058.77————
HIMRDTarget Model=Gemini-2.5-FT2026.028.4———0.42
HADES (method)Target Model=Qwen3-VL-plus2026.058.4————
SI-AttackTarget Model=GPT-4o2026.028.1———0.55
VisCRATarget Model=GPT-5-Chat2026.027.69———0.39
VideoJailTarget Model=GPT-4o2026.057.47————
TVPITarget Model=GPT-4.1 mini, Model Type=Closed-source2026.077.29————