Research projects

AACL-IJCNLP 2026 · Published

PAIRSAFE

A refusal score can rank two prompts correctly and still misclassify the safe one. PAIRSAFE follows both decisions across 150 matched pairs, comparing scores, held-out thresholds and generated responses.

Selected results

Matched pairs across six categories
150
Qwen pairs correct at held-out thresholds
132 / 150
Qwen pairs correct under the generation rule
105 / 150
Read the results

Scope of the evidence

The generation check covers Qwen and uses archived rule-based labels. The release reproduces the main table from scores and identifies the final nested metadata audit as a transcription from the paper.

Technical reports

Publication entry

PAIRSAFE: Proxy-Resistant Paired Safety Evaluation for Refusal Boundaries

Citation

Copy text

Copy was unavailable. Select and copy the text below.