DANGERLENS: Probing Dangerous Representations Hidden by Safety Alignment
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.
ProofPaper ↗
Key points
- 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.
- Experiments on three Qwen1.5 scales show that checkpoints with prior harmful exposure generally retain greater hazardous concept accessibility after the same safety alignment, with the strongest separation in intermediate and later layers.
Sources (1)
- [1]DANGERLENS: Probing Dangerous Representations Hidden by Safety AlignmentarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 14, 01:10 AM
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.
Safety alignment teaches language models to refuse harmful requests, yet the internal accessibility of hazardous concepts can retain traces of earlier training.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 14, 2026SERO: Learning from Executable Oracle Continuations for Long-Horizon Agentic Reinforcement Learning
- Oct 13, 2026Run fast, private agents on-device with LFM2.5-24B-A2B
- Oct 13, 2026Gemini 4 Argon: Built to solve your hardest problems
- Oct 13, 2026black-forest-labs/flux-3-image-klein-comfy
- Oct 13, 2026Cowork-Gym: Synthesizing Interactive Work Environments at Scale
- Oct 12, 2026Alibaba Qwen Introduces QwenGyre: Elastic Reinforcement Learning RL for XLong-Horizon Agents