Geometry-Lite: Interpretable Safety Probing via Layer-Wise Margin Geometry
The paper titled 'Geometry-Lite: Interpretable Safety Probing via Layer-Wise Margin Geometry' introduces a new method for analyzing safety in large language models. It focuses on how safety evidence is formed across different layers and proposes a compact prompt-level probe to improve detection performance. The study reveals that safety evidence is primarily expressed through persistent boundary-position geometry rather than layer-to-layer motion signals.
- ▪The Geometry-Lite method maps each layer's final prompt-token representation to signed margins for safety analysis.
- ▪It improves upon single-layer probes while maintaining performance close to multi-layer score stacking.
- ▪The study shows that final margins and unsafe-side layer occupancy are crucial for detection performance.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Machine Learning arXiv:2605.20241 (cs) [Submitted on 18 May 2026] Title:Geometry-Lite: Interpretable Safety Probing via Layer-Wise Margin Geometry Authors:Woo Seob Sim, Yu Rang Park View a PDF of the paper titled Geometry-Lite: Interpretable Safety Probing via Layer-Wise Margin Geometry, by Woo Seob Sim and 1 other authors View PDF HTML (experimental) Abstract:Prompt-level safety probes for large language models use hidden-state representations to separate safe from unsafe prompts, but strong average detection performance does not explain the geometry of this separation.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
