List your company

Credential Exfiltration Detection on Single-turn Prompts (Qwen-7B) (held-out)

0.998AUROCCIFT
0.719280.791640.8640.93636Jun 2, 2026

Evaluation results

MethodLinks
CIFTPre-out=✓2026.060.998991.5
Semantic emb.Pre-out=✗2026.060.9799
LlamaGuardPre-out=✗2026.060.857318
FuzzyPre-out=✗2026.060.86211
SubstringPre-out=✗2026.060.73518