Harmful SFT Leaves a Continuous Trace in LLM Checkpoint Updates

arXiv:2610.07518 · cs.LG, cs.AI, cs.CR · Submitted 2026-10-05 · Read on arXiv

cs.LG, cs.AI, cs.CR

Submitted: 2026-10-05

Updated: 2026-10-05

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

Terminology

• Trace — 1× in this paper · explained in 2 episodes

Sources

• Program Synthesis with Large Language Models
• Training Verifiers to Solve Math Word Problems
• The Llama 3 Herd of Models
• Covert Malicious Finetuning: Challenges in Safeguarding LLM Adaptation
• W2T: LoRA Weights Already Know What They Can Do
• Editing Models with Task Arithmetic
• Mistral 7B
• Insights from the Inverse: Reconstructing LLM Training Goals Through Inverse Reinforcement Learning
• Tulu 3: Pushing Frontiers in Open Language Model Post-Training
• Auditing language models for hidden objectives
• HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusal
• Olmo 3
• CANARY: Zero-Label Detection of Fine-Tuning Contamination in Language Models
• Spectral Geometry of LoRA Adapters Encodes Training Objective and Predicts Harmful Compliance
• AdvPrompter: Fast Adaptive Adversarial Prompting for LLMs
• Qwen3 Technical Report
• Universal and Transferable Adversarial Attacks on Aligned Language Models

Related papers

• Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
• AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
• Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
• Convergence issues in Relational Concept Analysis based on AOC-posets
• Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
• Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks