Multimodal Reward Hacking in Reinforcement Learning

• URL: http://arxiv.org/abs/2607.09492v1
• Date: Fri, 10 Jul 2026 15:06:48 GMT
• Title: Multimodal Reward Hacking in Reinforcement Learning
• Abstract summary: Reinforcement learning (RL) is increasingly used to align multimodal large language models (MLLMs)<n>We study reward hacking in MLLM RL across safety VQA, chart VQA, and stress-test settings.<n>We introduce Newly Rewarded Failure Rate (NRFR), which measures failures among samples whose proxy reward improves.
• Score: 25.776239829876065
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Reinforcement learning (RL) is increasingly used to align multimodal large language models (MLLMs), but higher rewards do not always imply better task performance. This risk is amplified when visual evidence is evaluated by text-only or weakly grounded rewards. We study reward hacking in MLLM RL across safety VQA, chart VQA, and stress-test settings, varying reward design, data ambiguity, model scale (2B-32B), and RL algorithm (GRPO, RLOO, DAPO). We introduce Newly Rewarded Failure Rate (NRFR), which measures failures among samples whose proxy reward improves over the SFT baseline. Outcome-only rewards cause severe hacking, reaching 48.1% Reward Hacking Rate (RHR), while NRFR exceeding RHR shows that RL creates new failures rather than merely inheriting them. Scaling reduces but does not eliminate hacking: even the 32B model retains a 54.9% worse rate under outcome-only rewards, whereas answer-aware rewards improve the oracle trend at every scale. Robustness is also algorithm- and scale-dependent: GRPO is consistently most resistant, RLOO remains vulnerable, and DAPO improves substantially from 2B to 8B. Visual-evidence rewards help only with reliable verification: keyword-based checks increase hacking, while VLM-as-judge semantic verification reduces it. Overall, multimodal reward hacking is a systematic result of optimizing imperfect rewards, and robust alignment requires rewards and verifiers that remain reliable under optimization pressure.

Related papers

• Two is better than one: A Collapse-free Multi-Reward RLIF Training Framework [6.490241400619907]
  Reinforcement learning from internal feedback has recently emerged as a scalable unsupervised alternative.<n>We propose a multi-reward RLIF framework that decomposes the training signal into two complementary components.<n>We show that our method can support stable long-horizon reasoning without relying on external ground-truth supervision.
  arXiv  Detail & Related papers  (2026-05-21T15:30:47Z)
• Mitigating Reward Hacking in RLHF via Advantage Sign Robustness [40.66292181512935]
  We make the assumption that reward hacking is often caused by flipped advantage signs.<n>By considering an adversarial perturbation in the RM parameter space, we can derive a certified sign-preservation radius.<n>We propose Sign-Certified Policy Optimization (SignCert-PO), down-weighting non-robust completions in the policy gradient update.
  arXiv  Detail & Related papers  (2026-04-03T11:45:16Z)
• IR$^3$: Contrastive Inverse Reinforcement Learning for Interpretable Detection and Mitigation of Reward Hacking [67.20568716300272]
  Reinforcement Learning from Human Feedback (RLHF) enables powerful LLM alignment but can introduce reward hacking.<n>We introduce IR3 (Interpretable Reward Reconstruction and Rectification), a framework that reverse-engineers, interprets, and surgically repairs the implicit objectives driving RLHF-tuned models.<n>We show that IR3 achieves 0.89 correlation with ground-truth rewards, identifies hacking features with over 90% precision, and significantly reduces hacking behaviors while maintaining capabilities within 3% of the original model.
  arXiv  Detail & Related papers  (2026-02-23T01:14:53Z)
• Reward Under Attack: Analyzing the Robustness and Hackability of Process Reward Models [68.45272703833209]
  We show that state-of-the-art PRMs are systematically exploitable under adversarial optimization pressure.<n>We introduce a three-tiered diagnostic framework that applies increasing adversarial pressure to quantify these vulnerabilities.<n>We release PRM-BiasBench and a diagnostic toolkit to enable robustness evaluation before deployment.
  arXiv  Detail & Related papers  (2026-02-20T23:38:03Z)
• Adversarial Reward Auditing for Active Detection and Mitigation of Reward Hacking [69.06218054848803]
  We propose Adrial Reward Auditing (ARA), a framework that reconceptualizes reward hacking as a dynamic, competitive game.<n>ARA operates in two stages: first, a Hacker policy discovers reward model vulnerabilities while an Auditor learns to detect exploitation from latent representations.<n>ARA achieves the best alignment-utility tradeoff among all baselines.
  arXiv  Detail & Related papers  (2026-02-02T07:34:57Z)
• From Verifiable Dot to Reward Chain: Harnessing Verifiable Reference-based Rewards for Reinforcement Learning of Open-ended Generation [52.62655622099456]
  We propose reinforcement learning with verifiable reference-based rewards (RLVRR)<n>Instead of checking the final answer, RLVRR extracts an ordered linguistic signal from high-quality references (i.e., reward chain)<n>In this way, RLVRR decomposes rewards into two dimensions: content, which preserves deterministic core concepts, and style, which evaluates adherence to stylistic properties.
  arXiv  Detail & Related papers  (2026-01-26T14:39:58Z)
• GDRO: Group-level Reward Post-training Suitable for Diffusion Models [55.948229011478304]
  Group-level rewards successfully align the model with the targeted reward.<n>Group-level Direct Reward Optimization (GDRO) is a new post-training paradigm for group-level reward alignment.<n>GDRO supports full offline training that saves the large time cost for image rollout sampling.<n>It is diffusion-sampler-independent, which eliminates the need for the ODE-to-SDE approximation to obtainity.
  arXiv  Detail & Related papers  (2026-01-05T11:47:18Z)
• Information-Theoretic Reward Modeling for Stable RLHF: Detecting and Mitigating Reward Hacking [78.69179041551014]
  We propose an information-theoretic reward modeling framework based on the Information Bottleneck principle.<n>We show that InfoRM filters out preference-irrelevant information to alleviate reward misgeneralization.<n>We also introduce IBL, a distribution-level regularization that penalizes such deviations, effectively expanding the optimization landscape.
  arXiv  Detail & Related papers  (2025-10-15T15:51:59Z)
• Stabilizing Long-term Multi-turn Reinforcement Learning with Gated Rewards [13.70228195630989]
  We introduce the SWE-oriented RL Framework, a unified system supporting multi-turn interaction, docker-based execution, and customizable reward functions.<n>We also propose Gated Reward Accumulation (G-RA), a novel method that accumulates immediate rewards only when high-level (long-term) rewards meet a predefined threshold.
  arXiv  Detail & Related papers  (2025-08-14T11:37:02Z)
• Stop Summation: Min-Form Credit Assignment Is All Process Reward Model Needs for Reasoning [36.320502608626896]
  Process reward models (PRMs) have proven effective for test-time scaling of Large Language Models (LLMs) on challenging reasoning tasks.<n>However, reward hacking issues with PRMs limit their successful application in reinforcement fine-tuning.<n>In this paper, we identify the main cause of PRM-induced reward hacking: the canonical summation-form credit assignment in reinforcement learning.
  arXiv  Detail & Related papers  (2025-04-21T17:59:02Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.