Multimodal Reward Hacking in Reinforcement Learning
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.
Read the paper
Related studies
- Meta-analysis and the science of research synthesis
- Rapid and sustained symptom reduction following psilocybin treatment for anxiety and depression in patients with life-threatening cancer: a randomized controlled trial
- Cluster Randomized Controlled Trial
- Convalescent plasma in patients admitted to hospital with COVID-19 (RECOVERY): a randomised controlled, open-label, platform trial
- Disease Free Survival and Local Recurrence for Laparoscopic Resection Compared to Open Resection of Stage II-III Rectal Cancer: Follow up Results of the ACOSOG Z6051 Randomized Controlled Trial
- The effects of a course of intranasal oxytocin on social behaviors in youth diagnosed with autism spectrum disorders: a randomized controlled trial
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.