AGENTIC-SYSTEMSJuly 23, 20261 min read

Geometry-Guided Constraint Learning for LLM Safety Classification

Researchers introduce a geometry-guided constraint learning approach using Sparse Autoencoders (SAEs) to streamline LLM safety classification, achieving 96-99% accuracy on Qwen3.5-9B.

Computer Science > Artificial Intelligence

Title:Geometry-Guided Constraint Learning for LLM Safety Classification

View PDF HTML (experimental)Abstract: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

Bibliographic and Citation Tools

Code, Data and Media Associated with this Article

Demos

Recommenders and Search Tools

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.

Source: arXiv cs.AI Recent

More in this category

agentic-systems

Rethinking Uncertainty Evaluation in Large Language Models

Traditional calibration is insufficient for evaluating LLM confidence as models frequently violate logical consistency despite appearing well-calibrated. Researchers propose the C1 framework along three axes—structural coherence, faithfulness, and usefulness—to bridge this gap.

agentic-systems

GraphContainer: A Unified Platform for Comparing and Debugging Graph RAG Methods

GraphContainer is a novel platform designed to unify and visualize diverse graph RAG workflows for LLMs. It features a Unified Graph Representation (UGR) layer and a Graph Recorder for live, traceable visual debugging and controlled comparison of graph retrieval strategies.

agentic-systems

Profile-Graph Memory for LLM Agents: Implicit Cross-Entity Traversal through Narrative Profiles

Researchers introduce ProGraph, a novel two-layer memory architecture designed to enhance multi-hop reasoning and precise recall for LLM agents. Alongside ProGraph, the team released MemHop, a new benchmark tailored for evaluating complex, multi-step memory retrieval.

agentic-systems

NEXUS: Structured Runtime Safety for Tool-Using LLM Agents

NEXUS is a novel structured-plan safety monitor designed for tool-using LLM agents. By combining deterministic rules and calibrated risk scoring, NEXUS enforces graded intervention policies with sub-millisecond latency.

agentic-systems

Information Discernment in Large Language Models

A new study introducing the Learn2Discern (L2D) benchmark reveals that large language models struggle to evaluate source reliability and truthfulness when integrating external knowledge. Researchers found that models often prioritize source popularity over reliability, though simple inference-time interventions can help mitigate these issues.

agentic-systems

OpenEvoShield: Dual Non-Stationary Continual Defense for Open-World Multi-Agent System Attacks

OpenEvoShield is a co-evolutionary continual defense framework designed to protect LLM-based multi-agent systems from dynamic injection attacks in open-world environments by addressing both fast attack evolution and slow normal behavior drift.