# CompassDPO: Dynamics-Controlled Direct Preference Optimization for Robust Safety Alignment

arXiv: https://arxiv.org/abs/2603.07211 · HTML: https://mlsanity.com/paper/2603.07211/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Placed in Safety alignment and jailbreak attacks, one of 10 directions in LLM alignment / RLHF / preference optimisation.
- Builds on 20 of its 33 references that are in the map.
- First posted on arXiv 7 Mar 2026.

## Placed in

- LLM alignment / RLHF / preference optimisation / [Safety alignment and jailbreak attacks](https://mlsanity.com/alignment/safety-alignment-and-jailbreak-attacks/)

## Builds on (most cited first, 20 of 20)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Proximal Policy Optimization Algorithms](https://mlsanity.com/paper/1707.06347/) (2017)
- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Deep reinforcement learning from human preferences](https://mlsanity.com/paper/1706.03741/) (2017)
- [MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmark](https://mlsanity.com/paper/2406.01574/) (2024)
- [SimPO: Simple Preference Optimization with a Reference-Free Reward](https://mlsanity.com/paper/2405.14734/) (2024)
- [A General Theoretical Paradigm to Understand Learning from Human Preferences](https://mlsanity.com/paper/2310.12036/) (2023)
- [BeaverTails: Towards Improved Safety Alignment of LLM via a Human-Preference Dataset](https://mlsanity.com/paper/2307.04657/) (2023)
- [ORPO: Monolithic Preference Optimization without Reference Model](https://mlsanity.com/paper/2403.07691/) (2024)
- [SLiC-HF: Sequence Likelihood Calibration with Human Feedback](https://mlsanity.com/paper/2305.10425/) (2023)
- [SALAD-Bench: A Hierarchical and Comprehensive Safety Benchmark for Large Language Models](https://mlsanity.com/paper/2402.05044/) (2024)
- [Do-Not-Answer: A Dataset for Evaluating Safeguards in LLMs](https://mlsanity.com/paper/2308.13387/) (2023)
- [Provably Robust DPO: Aligning Language Models with Noisy Feedback](https://mlsanity.com/paper/2403.00409/) (2024)
- [Direct Preference Optimization with an Offset](https://mlsanity.com/paper/2402.10571/) (2024)
- [Cal-DPO: Calibrated Direct Preference Optimization for Language Model Alignment](https://mlsanity.com/paper/2412.14516/) (2024)
- [Towards Data-Centric RLHF: Simple Metrics for Preference Dataset Comparison](https://mlsanity.com/paper/2409.09603/) (2024)
- [Robust Preference Optimization via Dynamic Target Margins](https://mlsanity.com/paper/2506.03690/) (2025)
- [Reward-Augmented Data Enhances Direct Preference Alignment of LLMs](https://mlsanity.com/paper/2410.08067/) (2024)
- [Revisiting Robustness for LLM Safety Alignment via Selective Geometry Control](https://mlsanity.com/paper/2602.07340/) (2026)

## Built on by (most cited first, 1 of 1)

- [Smaller Models, Better Rejects: Preference Distillation Scaling](https://mlsanity.com/paper/2609.38987/) (2026)

## Abstract (arXiv)

Direct Preference Optimization (DPO) has become a standard framework for safety alignment, but its reliance on pairwise preference updates makes training sensitive to imperfect supervision. Existing robust DPO methods often address this sensitivity through global loss corrections or external data-level interventions, while largely overlooking how unreliable comparisons distort batch-level optimization dynamics. We propose CompassDPO, a reward-free DPO framework that stabilizes preference optimization through dynamics control. Using the implicit DPO reward margin as a training-time compass, CompassDPO regulates sample influence along two complementary axes: update direction and update magnitude. For directional control, it applies sparse, budgeted, and warm-up delayed loss mixing to attenuate update components that conflict with the emerging preference direction. For magnitude control, it adaptively soft-winsorizes high-loss tail contributions, reducing tail dominance while preserving useful gradients from hard examples. Both mechanisms use only signals available during standard DPO training and require no external reward model or additional supervision. Experiments on PKU-SafeRLHF across four backbones and multiple out-of-distribution safety benchmarks show that CompassDPO consistently improves robustness over vanilla DPO and strong DPO-family baselines, especially under controlled label-flip noise. Code is available at https://anonymous.4open.science/r/CompassDPO-4D00