DANGERLENS: Probing Dangerous Representations Hidden by Safety Alignment
Safety alignment teaches language models to refuse harmful requests, yet the internal accessibility of hazardous concepts can retain traces of earlier training. Can checkpoints with equally safe behavior differ in what remains accessible behind their refusals? In this work, we investigate this question through paired training histories and a direct audit of intermediate representations. We introduce DANGERLENS, a probe that maps hidden states into vocabulary space and uses blinded semantic judgments to distinguish harmful goals and operational means from hazard recognition and safety control. Experiments on three Qwen1.5 scales show that checkpoints with prior harmful exposure generally r...
| Authors | Tianzhuo Yang, Shaoheng Yan, Lin Lu et al. |
|---|---|
| Primary category | cs.AI |
| Published | 14 October 2026 |
| Last updated at source | 14 October 2026 |
Silytics shows public metadata from arXiv and links to the original. Information may be incomplete or out of date. Last refreshed 15 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.