List your company

Targeted Attack on Image Captioning on Frontier MLLM Evaluation Set

89Attack Success Rate (ASR)M-Attack
-3.5620.4744.568.53May 15, 2026

Evaluation results

MethodLinks
M-AttackTarget MLLM=GPT-4.1, Variant=Ensemble, Attack Type=Sample-wise2026.058951
M-AttackTarget MLLM=Qwen3-VL-8B, Variant=Ensemble, Attack Type=Sample-wise2026.058747
M-AttackTarget MLLM=GPT-4o, Variant=Ensemble, Attack Type=Sample-wise2026.058647
M-AttackTarget MLLM=Gemini-2.5-Flash, Variant=Ensemble, Attack Type=Sample-wise2026.057738
M-AttackTarget MLLM=Gemini-3-Flash, Variant=Ensemble, Attack Type=Sample-wise2026.056932
DarkLLMTarget MLLM=GPT-4o, Variant=InternVL-2B, Attack Type=Universal2026.0567.728
DarkLLMTarget MLLM=GPT-4.1, Variant=InternVL-1B, Attack Type=Universal2026.0567.627
M-AttackTarget MLLM=GPT-4.1, Variant=L/14, Attack Type=Sample-wise2026.056631
M-AttackTarget MLLM=GPT-4o, Variant=L/14, Attack Type=Sample-wise2026.056531
DarkLLMTarget MLLM=Qwen3-VL-8B, Variant=InternVL-2B, Attack Type=Universal2026.0564.729
DarkLLMTarget MLLM=GPT-4o, Variant=InternVL-1B, Attack Type=Universal2026.0564.626
DarkLLMTarget MLLM=GPT-4.1, Variant=InternVL-2B, Attack Type=Universal2026.0564.627
DarkLLMTarget MLLM=Gemini-2.5-Flash, Variant=InternVL-2B, Attack Type=Universal2026.0563.625
M-AttackTarget MLLM=GPT-5-mini, Variant=Ensemble, Attack Type=Sample-wise2026.056329
DarkLLMTarget MLLM=Gemini-3-Flash, Variant=InternVL-2B, Attack Type=Universal2026.0560.623
AnyAttackTarget MLLM=GPT-4o, Variant=coco-bi, Attack Type=Universal2026.056022
AnyAttackTarget MLLM=GPT-4.1, Variant=coco-bi, Attack Type=Universal2026.055925
DarkLLMTarget MLLM=GPT-5-mini, Variant=InternVL-2B, Attack Type=Universal2026.0558.624
DarkLLMTarget MLLM=Gemini-2.5-Flash, Variant=InternVL-1B, Attack Type=Universal2026.0557.524
AnyAttackTarget MLLM=Qwen3-VL-8B, Variant=coco-bi, Attack Type=Universal2026.055624
DarkLLMTarget MLLM=GPT-5-mini, Variant=InternVL-1B, Attack Type=Universal2026.0554.524
M-AttackTarget MLLM=Gemini-2.5-Flash, Variant=L/14, Attack Type=Sample-wise2026.055325
AnyAttackTarget MLLM=GPT-5-mini, Variant=coco-bi, Attack Type=Universal2026.055322
DarkLLMTarget MLLM=Gemini-3-Flash, Variant=InternVL-1B, Attack Type=Universal2026.0552.522
M-AttackTarget MLLM=Qwen3-VL-8B, Variant=L/14, Attack Type=Sample-wise2026.055228
M-AttackTarget MLLM=GPT-4.1, Variant=B/16, Attack Type=Sample-wise2026.055020
DarkLLMTarget MLLM=Qwen3-VL-8B, Variant=InternVL-1B, Attack Type=Universal2026.0549.524
AnyAttackTarget MLLM=GPT-4.1, Variant=flickr-bi, Attack Type=Universal2026.054719
AnyAttackTarget MLLM=Gemini-3-Flash, Variant=coco-bi, Attack Type=Universal2026.054618
AnyAttackTarget MLLM=Gemini-2.5-Flash, Variant=coco-bi, Attack Type=Universal2026.054419
M-AttackTarget MLLM=GPT-4o, Variant=B/16, Attack Type=Sample-wise2026.054319
M-AttackTarget MLLM=Gemini-3-Flash, Variant=L/14, Attack Type=Sample-wise2026.054217
AnyAttackTarget MLLM=GPT-4o, Variant=flickr-bi, Attack Type=Universal2026.054016
AnyAttackTarget MLLM=Qwen3-VL-8B, Variant=flickr-bi, Attack Type=Universal2026.053617
M-AttackTarget MLLM=GPT-5-mini, Variant=L/14, Attack Type=Sample-wise2026.053416
AnyAttackTarget MLLM=GPT-4.1, Variant=flickr-cos, Attack Type=Universal2026.053414
AnyAttackTarget MLLM=GPT-5-mini, Variant=flickr-bi, Attack Type=Universal2026.053216
AnyAttackTarget MLLM=Gemini-2.5-Flash, Variant=flickr-bi, Attack Type=Universal2026.053214
M-AttackTarget MLLM=Qwen3-VL-8B, Variant=B/16, Attack Type=Sample-wise2026.053115
AnyAttackTarget MLLM=Gemini-2.5-Flash, Variant=flickr-cos, Attack Type=Universal2026.053014
AnyAttackTarget MLLM=GPT-4o, Variant=flickr-cos, Attack Type=Universal2026.052914
AnyAttackTarget MLLM=GPT-5-mini, Variant=flickr-cos, Attack Type=Universal2026.052712
M-AttackTarget MLLM=GPT-5-mini, Variant=B/16, Attack Type=Sample-wise2026.052612
M-AttackTarget MLLM=Gemini-2.5-Flash, Variant=B/16, Attack Type=Sample-wise2026.052511
AnyAttackTarget MLLM=Qwen3-VL-8B, Variant=flickr-cos, Attack Type=Universal2026.052511
AnyAttackTarget MLLM=Gemini-3-Flash, Variant=flickr-bi, Attack Type=Universal2026.05239
M-AttackTarget MLLM=Gemini-3-Flash, Variant=B/16, Attack Type=Sample-wise2026.05218
AnyAttackTarget MLLM=Gemini-3-Flash, Variant=flickr-cos, Attack Type=Universal2026.05209
AnyAttackTarget MLLM=Gemini-3-Flash, Variant=coco-cos, Attack Type=Universal2026.0583
AnyAttackTarget MLLM=GPT-4o, Variant=coco-cos, Attack Type=Universal2026.0574
AnyAttackTarget MLLM=Gemini-2.5-Flash, Variant=coco-cos, Attack Type=Universal2026.0574
AnyAttackTarget MLLM=Qwen3-VL-8B, Variant=coco-cos, Attack Type=Universal2026.0552
AnyAttackTarget MLLM=GPT-5-mini, Variant=coco-cos, Attack Type=Universal2026.0554
AttackVLMTarget MLLM=Qwen3-VL-8B, Variant=L/14, Attack Type=Sample-wise2026.0543
AnyAttackTarget MLLM=GPT-4.1, Variant=coco-cos, Attack Type=Universal2026.0543
AttackVLMTarget MLLM=Qwen3-VL-8B, Variant=B/16, Attack Type=Sample-wise2026.0532
AttackVLMTarget MLLM=GPT-4o, Variant=B/16, Attack Type=Sample-wise2026.0531
AttackVLMTarget MLLM=GPT-4.1, Variant=B/16, Attack Type=Sample-wise2026.0532
AttackVLMTarget MLLM=GPT-4o, Variant=L/14, Attack Type=Sample-wise2026.0532
AttackVLMTarget MLLM=Qwen3-VL-8B, Variant=Ensemble, Attack Type=Sample-wise2026.0532
AttackVLMTarget MLLM=GPT-4.1, Variant=Ensemble, Attack Type=Sample-wise2026.0531
AttackVLMTarget MLLM=GPT-5-mini, Variant=Ensemble, Attack Type=Sample-wise2026.0531
AttackVLMTarget MLLM=Gemini-2.5-Flash, Variant=L/14, Attack Type=Sample-wise2026.0522
AttackVLMTarget MLLM=GPT-4o, Variant=Ensemble, Attack Type=Sample-wise2026.0522
AttackVLMTarget MLLM=Gemini-2.5-Flash, Variant=Ensemble, Attack Type=Sample-wise2026.0522
AttackVLMTarget MLLM=Gemini-3-Flash, Variant=Ensemble, Attack Type=Sample-wise2026.0521
AttackVLMTarget MLLM=Gemini-2.5-Flash, Variant=B/16, Attack Type=Sample-wise2026.0511
AttackVLMTarget MLLM=GPT-4.1, Variant=L/14, Attack Type=Sample-wise2026.0511
AttackVLMTarget MLLM=GPT-5-mini, Variant=L/14, Attack Type=Sample-wise2026.0510
AttackVLMTarget MLLM=Gemini-3-Flash, Variant=L/14, Attack Type=Sample-wise2026.0510
AttackVLMTarget MLLM=GPT-5-mini, Variant=B/16, Attack Type=Sample-wise2026.0500
AttackVLMTarget MLLM=Gemini-3-Flash, Variant=B/16, Attack Type=Sample-wise2026.0500