List your company

Jailbreak Detection on I-FSJ

81Transferred Detection Rateembedding disruption method
67.4870.9974.578.01May 11, 2026

Evaluation results

MethodLinks
embedding disruption methodModels=Qwen2.5-7B-Instruct2026.0581
embedding disruption methodModels=LLaMA2-7B-Chat2026.0579
embedding disruption methodModels=Gemini-2.52026.0579
embedding disruption methodModels=Vicuna-13B2026.0574
embedding disruption methodModels=GPT-4.12026.0568