REIN: Bridging the Gap between Reasoning and Reliability via Reflection and Abstention Alignment

2026-08-08

AI Summary

This work addresses the unreliability of large reasoning models, which often suffer from reasoning errors or knowledge hallucinations. To mitigate these issues, the authors propose REIN, a novel alignment framework that uniquely integrates explicit self-reflection with an active abstention mechanism. Within a single forward pass and without requiring process supervision or external retrieval, REIN generates structured reasoning chains following a “think → reflect → respond” paradigm and proactively abstains when confident evidence is lacking. The approach substantially reduces both types of hallucinations, achieving a 58%–72% reduction in hallucination rates compared to baseline methods across multiple benchmarks, while maintaining high coverage (86%–91%). Furthermore, among answered samples, it improves selective accuracy by 6.6–14.2%.

Abstract

Large reasoning models (LRMs) are prone to hallucination, which undermines their reliability and poses challenges for safe deployment. Hallucinations in LRMs arise from two distinct failure sources: reasoning hallucination, where flawed inference steps propagate to an incorrect conclusion, and knowledge hallucination, where the model lacks the requisite factual knowledge to answer the query. To address reasoning hallucination, we propose REIN, an alignment framework that trains LRMs to produce a structured reasoning sequence, $\texttt{<think>} $$\rightarrow$ $\texttt{<reflection>} $$\rightarrow$ $\texttt{<answer>}$, enabling explicit self-reflection before committing to a final answer. To address knowledge hallucination, REIN introduces a reward mechanism that encourages explicit abstention (e.g., "I don't know") when none of the sampled reasoning chains yields a correct answer, allowing the model to refrain from unsupported predictions. Extensive evaluations on mathematical and commonsense reasoning benchmarks show that REIN consistently improves selective accuracy, reduces incorrect-but-self-endorsed responses, and maintains high coverage compared with competitive baselines. Notably, REIN achieves these gains within a single forward pass, without requiring process supervision, inference-time controllers, external search, or multi-round critiques. Experiments on multiple backbones show that REIN reduces the hallucination proxy by $58\sim72\%$ relative to the base models while maintaining $86\sim91\%$ average coverage, and improves selective accuracy on attempted questions by $6.6\sim14.2\%$.

Problem

• hallucination
• reasoning
• reliability
• large reasoning models
• abstention

Innovation

• reasoning hallucination
• knowledge hallucination
• self-reflection
• abstention alignment
• selective accuracy