ResearchBenchmarksQuestion Answering on OpenAI APIs TwoOptions MaliciousQuestion Answering on OpenAI APIs TwoOptions MaliciousFollow84.44F1 ScorePARASITE54.664862.394970.12577.8551May 22, 2025Evaluation resultsMethodMethodLinksF1 ScoreExact Match (EM)PARASITEModel=Qwen2.5-7BModel=Qwen2.5-7B2025.0584.4480PARASITEModel=Deepseek-7BModel=Deepseek-7B2025.0555.8142.5