List your company

Prompt recovery attack on CounterFact

60Top-1 AccuracyPrompt Recovery Attack
18.429.24050.8Feb 7, 2026

Evaluation results

MethodLinks
Prompt Recovery AttackModel=GPT-J, Editing Algorithm=ROME2026.02608010086
Prompt Recovery AttackModel=Llama-3-Instruct, Editing Algorithm=ROME2026.02608010089
KSTERModel=GPT-J, N=10, Target Editing Method=MEMIT2026.0257869688
KSTERModel=Llama-3-Instruct, N=10, Target Editing Method=MEMIT2026.0253909688
KSTERModel=GPT-J, N=50, Target Editing Method=MEMIT2026.0251909988
KSTERModel=Llama-3-Instruct, N=100, Target Editing Method=MEMIT2026.0251819488
KSTERModel=GPT-J, N=100, Target Editing Method=MEMIT2026.0250929988
KSTERModel=Llama-3-Instruct, N=50, Target Editing Method=MEMIT2026.0249829488
KSTERModel=Qwen-2.5-Instruct, N=100, Target Editing Method=MEMIT2026.0247869786
KSTERModel=Qwen-2.5-Instruct, N=50, Target Editing Method=MEMIT2026.0242819586
KSTERModel=Qwen-2.5-Instruct, N=10, Target Editing Method=MEMIT2026.0235839685
Prompt Recovery AttackModel=Qwen-2.5-Instruct, Editing Algorithm=ROME2026.02204010084