List your company

Workflow Steering Attack on MisinfoTask 108 constrained analytical tasks

75TASRFLOWSTEER
9.521626.520843.5260.5192May 12, 2026

Evaluation results

MethodLinks
FLOWSTEERModel Family=Gemini-2.5, Planner-Executor Setup=G22026.057575
FLOWSTEERModel Family=GPT-4o, Planner-Executor Setup=G1 (Same LLM for planner and executors)2026.0563.8965.74
FLOWSTEERModel Family=Gemini-2.5, Planner-Executor Setup=G12026.0563.8963.89
FLOWSTEERModel Family=GPT-4o, Planner-Executor Setup=G2 (Strengthens the planner)2026.0558.3361.11
Direct Goal Injection (DGI)Model Family=Gemini-2.5, Planner-Executor Setup=G12026.0551.8549.07
FLOWSTEERModel Family=Qwen-3.5, Planner-Executor Setup=G22026.0551.8557.41
FLOWSTEERModel Family=DeepSeek, Planner-Executor Setup=G12026.0551.8556.48
FLOWSTEERModel Family=DeepSeek, Planner-Executor Setup=G32026.0549.0750
FLOWSTEERModel Family=DeepSeek, Planner-Executor Setup=G22026.0547.2254.63
FLOWSTEERModel Family=Qwen-3.5, Planner-Executor Setup=G12026.0546.351.85
FLOWSTEERModel Family=Gemini-2.5, Planner-Executor Setup=G32026.0544.4448.15
FLOWSTEERModel Family=GPT-4o, Planner-Executor Setup=G3 (Strengthens the executors)2026.0539.8148.15
FLOWSTEERModel Family=Qwen-3.5, Planner-Executor Setup=G32026.0538.8942.59
Direct Goal Injection (DGI)Model Family=GPT-4o, Planner-Executor Setup=G1 (Same LLM for planner and executors)2026.0533.3335.18
Naive Malicious Prompting (NM)Model Family=GPT-4o, Planner-Executor Setup=G1 (Same LLM for planner and executors)2026.0527.7829.63
Naive Malicious Prompting (NM)Model Family=GPT-4o, Planner-Executor Setup=G2 (Strengthens the planner)2026.052525.93
Naive Malicious Prompting (NM)Model Family=Gemini-2.5, Planner-Executor Setup=G12026.0524.0714.81
Direct Goal Injection (DGI)Model Family=GPT-4o, Planner-Executor Setup=G2 (Strengthens the planner)2026.0523.1529.63
Direct Goal Injection (DGI)Model Family=GPT-4o, Planner-Executor Setup=G3 (Strengthens the executors)2026.0517.5922.22
Naive Malicious Prompting (NM)Model Family=GPT-4o, Planner-Executor Setup=G3 (Strengthens the executors)2026.0512.0416.67