AI Paper Atlas

Research map / Adversarial robustness and security / Attacks on text and graphs

Reward Under Attack: Analyzing the Robustness and Hackability of Process Reward Models

Rishabh Tiwari, Aditya Tomar, Udbhav Bamba, Monishwaran Maheswaran, Heng Yang, Michael W. Mahoney, Kurt Keutzer, Amir Gholami

ICML 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: LLM jailbreak and injection defense · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
append 100 optimized adversarial tokensSkywork-o1-Open-PRM-1.5B · AIME25 transfermean PRM reward: 0.305 -> 0.924Improved
append 50 optimized adversarial tokensSkywork-o1-Open-PRM-1.5B · AIME25 transfermean PRM reward: +0.224Improved
append 50 optimized adversarial tokensSkywork-o1-Open-PRM-7B · AIME25 transfermean PRM reward: +0.070Improved
append 100 optimized adversarial tokensQwen2.5-Math-PRM-7B · AIME24 training rewardbest training reward: 0.658 -> 0.437Hurt
append 100 optimized adversarial tokensQwen2.5-Math-PRM-7B · AIME25 transferreward change: -0.042Hurt
increase adversarial tokens from 1 to 100Skywork-o1-Open-PRM-1.5B · AIME25 transferreward change: +0.030 -> +0.619Improved
use adversarial rather than random tokens (100 tokens)Skywork-o1-Open-PRM-1.5B · AIME24 reward landscapebasin volume: 1.495 vs 0.689Improved
use adversarial rather than random tokens (100 tokens)Qwen2.5-Math-PRM-7B · AIME24 reward landscapebasin volume: 1.570 vs 0.421Improved
train policy with GRPO on PRM rewardQwen2.5-1.5B-Instruct · AIME25mean PRM reward (Skywork-1.5B): 0.246 -> 0.641Improved
GRPO training on PRM rewardQwen2.5-1.5B-Instruct · AIME24ground-truth accuracy: near zero (peaking at 3-4%)No clear effect
GRPO training on Qwen-7B PRM rewardQwen2.5-1.5B-Instruct · AIME24ground-truth accuracy: accuracy drops to 0%Hurt
rephrase GRPO trajectories (semantics-preserving) (ablation)GRPO policy with Skywork-1.5B PRM · AIME25mean PRM reward: 0.641 -> 0.472Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy