ASCENT: First-Order Optimal Fine-Tuning with Recalibration for Safety--Utility Co-Enhancement

arXiv:2610.08061 · cs.CR · Submitted 2026-10-06 · Read on arXiv

cs.CR

Submitted: 2026-10-06

Updated: 2026-10-06

Code: https://github.com/ZJU-LLM-Safety/ASCENT

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Terminology

• Fine-Tuning — 1× in this paper · explained in 5 episodes

Sources

• Training Verifiers to Solve Math Word Problems
• The Llama 3 Herd of Models
• NonTextual Target Attack
• DualBreach: Efficient Dual-Jailbreaking via Target-Driven Initialization and Multi-Target Optimization
• HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusal
• DARWIN: Evolving Jailbreak Adversary and Guardrail for LLM Safety Evaluation and Protection
• Qwen2.5 Technical Report
• Gemma 3 Technical Report
• Assessing the Brittleness of Safety Alignment via Pruning and Low-Rank Modifications
• DataShield: Uncovering Risky Fine-Tuning Data Across LLMs Through Consensus Subspace Alignment
• Qwen3 Technical Report
• HarmMetric Eval: Benchmarking Metrics and Judges for LLM Harmfulness Assessment
• TRACE: Task-Aware Adaptive Self-Evolving Agentic Jailbreaking
• TrueMoE: Dual-Routing Mixture of Discriminative Experts for Synthetic Image Detection

Related papers

• SoK: AI-Augmented Binary Reversing
• Relaxed Sender Anonymity for CBDC Interbank Settlement: A Zero-Knowledge Approach on Permissioned EVM
• Calibration-Family Overfit: Why Trusted Sabotage Monitors Don't Transfer Across Lineages
• Efficient Fuzzy PSI under One-Sided Assumptions
• Sealing the Audit-Runtime Gap for LLM Skills
• Token Composition: A Graph Based on EVM Logs