SOURCE-LINKED INTELLIGENCE

Beyond Average Safety: Chance-Constrained LLM Fine-tuning

arXiv · AI, language, vision and robotics · article · Sep 24, 2026 · UTC

Fine-tuning large language models on new objectives can improve helpfulness, instruction following, or domain-specific performance, but it can also induce regressions on safety-critical prompts. Existing safety-preserving fine-tuning methods typically control average safety loss or use weighted auxiliary penalties, which can obscure rare but severe failures. We propose a chance-constrained formulation for safety-preserving fine-tuning that limits the fraction of safety examples whose degradation relative to a reference model exceeds a prescribed threshold. Because the resulting empirical chanc

Read original source ↗ Open in workspace

recordType

paper

region

Global