├── .gitignore ├── .gitmodules ├── .python-version ├── CODE_OF_CONDUCT.md ├── CONTRIBUTING.md ├── LICENSE ├── README.md ├── assets └── summary.png ├── data ├── README.md └── results │ └── single_prompt │ └── embed │ └── prompt0_image2 │ ├── generations │ ├── jbb_completions_train.json │ └── jbb_completions_train_shortcut.json │ ├── images │ ├── BEST │ ├── FINAL.pth │ ├── step100_auto │ └── step200_auto │ └── prompt0_image2.yaml ├── experiments ├── attacks │ ├── single_image │ │ ├── embed │ │ │ └── prompt0_image2.yaml │ │ └── vocab │ │ │ └── prompt0_image2_1prevent.yaml │ └── transfer │ │ ├── embed │ │ └── 10prompts.yaml │ │ └── vocab │ │ └── 10prompts.yaml └── train_shortcut │ ├── embed_cosine.yaml │ └── vocab.yaml ├── pyproject.toml ├── src └── fusion_jailbreaks │ ├── __init__.py │ ├── config.py │ ├── eval │ ├── __init__.py │ ├── compute_perplexity.py │ ├── eval_gcg.py │ ├── evaluation.py │ └── transferability_llava.py │ ├── run_attack.py │ ├── train.py │ └── utils.py └── uv.lock /.gitignore: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/.gitignore -------------------------------------------------------------------------------- /.gitmodules: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/.gitmodules -------------------------------------------------------------------------------- /.python-version: -------------------------------------------------------------------------------- 1 | cpython@3.11.9 2 | -------------------------------------------------------------------------------- /CODE_OF_CONDUCT.md: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/CODE_OF_CONDUCT.md -------------------------------------------------------------------------------- /CONTRIBUTING.md: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/CONTRIBUTING.md -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/LICENSE -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/README.md -------------------------------------------------------------------------------- /assets/summary.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/assets/summary.png -------------------------------------------------------------------------------- /data/README.md: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /data/results/single_prompt/embed/prompt0_image2/generations/jbb_completions_train.json: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/data/results/single_prompt/embed/prompt0_image2/generations/jbb_completions_train.json -------------------------------------------------------------------------------- /data/results/single_prompt/embed/prompt0_image2/generations/jbb_completions_train_shortcut.json: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/data/results/single_prompt/embed/prompt0_image2/generations/jbb_completions_train_shortcut.json -------------------------------------------------------------------------------- /data/results/single_prompt/embed/prompt0_image2/images/BEST: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/data/results/single_prompt/embed/prompt0_image2/images/BEST -------------------------------------------------------------------------------- /data/results/single_prompt/embed/prompt0_image2/images/FINAL.pth: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/data/results/single_prompt/embed/prompt0_image2/images/FINAL.pth -------------------------------------------------------------------------------- /data/results/single_prompt/embed/prompt0_image2/images/step100_auto: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/data/results/single_prompt/embed/prompt0_image2/images/step100_auto -------------------------------------------------------------------------------- /data/results/single_prompt/embed/prompt0_image2/images/step200_auto: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/data/results/single_prompt/embed/prompt0_image2/images/step200_auto -------------------------------------------------------------------------------- /data/results/single_prompt/embed/prompt0_image2/prompt0_image2.yaml: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/data/results/single_prompt/embed/prompt0_image2/prompt0_image2.yaml -------------------------------------------------------------------------------- /experiments/attacks/single_image/embed/prompt0_image2.yaml: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/experiments/attacks/single_image/embed/prompt0_image2.yaml -------------------------------------------------------------------------------- /experiments/attacks/single_image/vocab/prompt0_image2_1prevent.yaml: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/experiments/attacks/single_image/vocab/prompt0_image2_1prevent.yaml -------------------------------------------------------------------------------- /experiments/attacks/transfer/embed/10prompts.yaml: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/experiments/attacks/transfer/embed/10prompts.yaml -------------------------------------------------------------------------------- /experiments/attacks/transfer/vocab/10prompts.yaml: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/experiments/attacks/transfer/vocab/10prompts.yaml -------------------------------------------------------------------------------- /experiments/train_shortcut/embed_cosine.yaml: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/experiments/train_shortcut/embed_cosine.yaml -------------------------------------------------------------------------------- /experiments/train_shortcut/vocab.yaml: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/experiments/train_shortcut/vocab.yaml -------------------------------------------------------------------------------- /pyproject.toml: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/pyproject.toml -------------------------------------------------------------------------------- /src/fusion_jailbreaks/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/src/fusion_jailbreaks/__init__.py -------------------------------------------------------------------------------- /src/fusion_jailbreaks/config.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/src/fusion_jailbreaks/config.py -------------------------------------------------------------------------------- /src/fusion_jailbreaks/eval/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/src/fusion_jailbreaks/eval/__init__.py -------------------------------------------------------------------------------- /src/fusion_jailbreaks/eval/compute_perplexity.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/src/fusion_jailbreaks/eval/compute_perplexity.py -------------------------------------------------------------------------------- /src/fusion_jailbreaks/eval/eval_gcg.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/src/fusion_jailbreaks/eval/eval_gcg.py -------------------------------------------------------------------------------- /src/fusion_jailbreaks/eval/evaluation.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/src/fusion_jailbreaks/eval/evaluation.py -------------------------------------------------------------------------------- /src/fusion_jailbreaks/eval/transferability_llava.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/src/fusion_jailbreaks/eval/transferability_llava.py -------------------------------------------------------------------------------- /src/fusion_jailbreaks/run_attack.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/src/fusion_jailbreaks/run_attack.py -------------------------------------------------------------------------------- /src/fusion_jailbreaks/train.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/src/fusion_jailbreaks/train.py -------------------------------------------------------------------------------- /src/fusion_jailbreaks/utils.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/src/fusion_jailbreaks/utils.py -------------------------------------------------------------------------------- /uv.lock: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/facebookresearch/multimodal-fusion-jailbreaks/HEAD/uv.lock --------------------------------------------------------------------------------