List your company

Stateful Agent Backdoor Attack on Branch-and-merge instantiation 20 episodes

100R (Initial)Ministral-3-14B
7480.7587.594.25May 7, 2026

Evaluation results

MethodLinks
Ministral-3-14Binstantiation=branch-and-merge, episodes_per_model=202026.05100100100100100000
Llama-3.1-8Binstantiation=branch-and-merge, episodes_per_model=202026.0590908580850100
Qwen2.5-7Binstantiation=branch-and-merge, episodes_per_model=202026.059090859080050
Qwen2.5-14Binstantiation=branch-and-merge, episodes_per_model=202026.0575705565700350