{"slug": "geometry-guided-constraint-learning-for-llm-safety-classification", "title": "Geometry-Guided Constraint Learning for LLM Safety Classification", "summary": "A new method called Geometry-Guided Constraint Learning improves LLM safety classification by using sparse autoencoder (SAE) feature extraction, achieving 96-99% accuracy per category on the BeaverTails benchmark with Qwen3.5-9B. The approach reduces the optimal constraint count K to 2 for 12 out of 14 categories, eliminating the need for exhaustive parameter sweeps. The authors introduce a cone constraint with learnable aperture to adapt to each category's cluster concentration.", "body_md": "arXiv:2607.19366v1 Announce Type: new\nAbstract: Safety as Polytope (SaP) learns linear half-space constraints in LLM hidden space but requires per-category tuning of the constraint count K. We show that sparse autoencoder (SAE) feature extraction resolves this: K=2 becomes optimal for 12/14 categories on Qwen3.5-9B, achieving 96-99% accuracy per category on our BeaverTails classification benchmark, largely eliminating the need for exhaustive sweeps (K=4-25 with random initialization). This convergence to two planes is consistent with the Linear Representation Hypothesis, providing suggestive evidence that safety boundaries in this setting admit a low-dimensional linear description in the SAE feature space. Building on this geometric perspective, we introduce a cone constraint whose learnable aperture adapts to each category's cluster concentration, stabilized by a three-phase training", "url": "https://wpnews.pro/news/geometry-guided-constraint-learning-for-llm-safety-classification", "canonical_source": "https://arxiv.org/abs/2607.19366", "published_at": "2026-07-23 04:00:00+00:00", "updated_at": "2026-07-23 04:09:01.829783+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-safety", "ai-research"], "entities": ["Qwen3.5-9B", "BeaverTails", "arXiv", "SaP", "SAE"], "alternates": {"html": "https://wpnews.pro/news/geometry-guided-constraint-learning-for-llm-safety-classification", "markdown": "https://wpnews.pro/news/geometry-guided-constraint-learning-for-llm-safety-classification.md", "text": "https://wpnews.pro/news/geometry-guided-constraint-learning-for-llm-safety-classification.txt", "jsonld": "https://wpnews.pro/news/geometry-guided-constraint-learning-for-llm-safety-classification.jsonld"}}