# Learning from Failures: Understanding LLM Alignment through Failure-Aware Inverse RL

arXiv: https://arxiv.org/abs/2510.06092 · HTML: https://mlsanity.com/paper/2510.06092/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Placed in Reward modeling and RLHF, one of 10 directions in LLM alignment / RLHF / preference optimisation.
- Builds on 23 of its 32 references that are in the map.
- First posted on arXiv 7 Oct 2025.

## Placed in

- LLM alignment / RLHF / preference optimisation / [Reward modeling and RLHF](https://mlsanity.com/alignment/reward-modeling-and-rlhf/)

## Builds on (most cited first, 23 of 23)

- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [GPT-4 Technical Report](https://mlsanity.com/paper/2303.08774/) (2023)
- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Explaining and Harnessing Adversarial Examples](https://mlsanity.com/paper/1412.6572/) (2014)
- [On the Opportunities and Risks of Foundation Models](https://mlsanity.com/paper/2108.07258/) (2021)
- [Deep reinforcement learning from human preferences](https://mlsanity.com/paper/1706.03741/) (2017)
- [Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling](https://mlsanity.com/paper/2304.01373/) (2023)
- [RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Models](https://mlsanity.com/paper/2009.11462/) (2020)
- [MentorNet: Learning Data-Driven Curriculum for Very Deep Neural Networks on Corrupted Labels](https://mlsanity.com/paper/1712.05055/) (2017)
- [Guided Cost Learning: Deep Inverse Optimal Control via Policy Optimization](https://mlsanity.com/paper/1603.00448/) (2016)
- [Not All Samples Are Created Equal: Deep Learning with Importance Sampling](https://mlsanity.com/paper/1803.00942/) (2018)
- [Extrapolating Beyond Suboptimal Demonstrations via Inverse Reinforcement Learning from Observations](https://mlsanity.com/paper/1904.06387/) (2019)
- [Inverse Reward Design](https://mlsanity.com/paper/1711.02827/) (2017)
- [Learning from Suboptimal Demonstration via Self-Supervised Reward Regression](https://mlsanity.com/paper/2010.11723/) (2020)
- [Invariance in Policy Optimisation and Partial Identifiability in Reward Learning](https://mlsanity.com/paper/2203.07475/) (2022)
- [Towards Resolving Unidentifiability in Inverse Reinforcement Learning](https://mlsanity.com/paper/1601.06569/) (2016)
- [STARC: A General Framework For Quantifying Differences Between Reward Functions](https://mlsanity.com/paper/2309.15257/) (2023)
- [Reward Model Overoptimisation in Iterated RLHF](https://mlsanity.com/paper/2505.18126/) (2025)
- [Inverse Reinforcement Learning Meets Large Language Model Post-Training: Basics, Advances, and Opportunities](https://mlsanity.com/paper/2507.13158/) (2025)
- [Insights from the Inverse: Reconstructing LLM Training Goals Through Inverse Reinforcement Learning](https://mlsanity.com/paper/2410.12491/) (2024)
- [Learning a Pessimistic Reward Model in RLHF](https://mlsanity.com/paper/2505.20556/) (2025)
- [Partial Identifiability and Misspecification in Inverse Reinforcement Learning](https://mlsanity.com/paper/2411.15951/) (2024)
- [Teach a Reward Model to Correct Itself: Reward Guided Adversarial Failure Discovery for Robust Reward Modeling](https://mlsanity.com/paper/2507.06419/) (2025)

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

- [Are Aligned Large Language Models Still Misaligned?](https://mlsanity.com/paper/2602.11305/) (2026)

## Abstract (arXiv)

Reinforcement Learning from Human Feedback (RLHF) aligns Large Language Models (LLMs) with human preferences, yet the underlying reward signals they internalize remain hidden, posing a critical challenge for interpretability and safety. Existing approaches attempt to extract these latent incentives using Inverse Reinforcement Learning (IRL), but treat all preference pairs equally, often overlooking the most informative signals: those examples the extracted reward model misclassifies or assigns nearly equal scores, which we term failures. We introduce a novel failure-aware IRL algorithm that focuses on misclassified or difficult examples to recover the latent rewards defining model behaviors. By learning from these failures, our failure-aware IRL extracts reward functions that better reflect the true objectives behind RLHF. We demonstrate that failure-aware IRL outperforms existing IRL baselines across multiple metrics when applied to LLM detoxification, without requiring external classifiers or supervision. Crucially, failure-aware IRL yields rewards that better capture the true incentives learned during RLHF, enabling more effective re-RLHF training than standard IRL. This establishes failure-aware IRL as a robust, scalable method for auditing model alignment and reducing ambiguity in the IRL process.