Right Answer, Wrong Method: Shortcut Hacking Misleads the Evaluation of LLM Reasoning on Frontier Science Benchmarks
Xuan Ren, Weiqi Zhai, Tianle Pu, Yihua Zhu, Yihua Zhu, Hu Wei, Bing Zhao
cs.AI, cs.CL
Submitted: 2026-08-03
Comments: working in progress
License: http://creativecommons.org/licenses/by/4.0/
The gist: Scientific reasoning benchmarks typically evaluate large language models (LLMs) using final-answer accuracy.
Terminology
Abstract
Scientific reasoning benchmarks typically evaluate large language models (LLMs) using final-answer accuracy. However, a correct answer does not necessarily demonstrate the reasoning capability targeted by the problem. We identify Solution Hacking, a failure mode in which an LLM reaches the correct answer through invalid shortcuts, such as numerical search, enumeration, guessing, or answer-first verification, without providing a valid task-targeted derivation. We systematically analyze this phenomenon across difficulty levels, scientific domains, and frontier models. Solution hacking increases sharply with benchmark difficulty, from 2.2% on common problems to 28.3% on Olympiad-level problems and 37.4% on HLE. Moreover, 8.2%-44.1% of answers credited as correct across frontier models are identified as hacked solutions. We further develop expert-inspired anti-hacking strategies, including an automatic judge and a test-time instruction. The results show that suppressing shortcut behavior substantially reduces reported accuracy while having a smaller effect on correct and non-hacked accuracy. These findings reveal that answer-only evaluation can overestimate the scientific reasoning capabilities of frontier LLMs.
Sources
- Humanity's Last Exam
- Omni-MATH: A Universal Olympiad Level Mathematic Benchmark For Large Language Models
- PHYBench: Holistic Evaluation of Physical Perception and Reasoning in Large Language Models
- Sycophancy to Subterfuge: Investigating Reward-Tampering in Large Language Models
- Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation
- Don't Make Your LLM an Evaluation Benchmark Cheater
- A Careful Examination of Large Language Model Performance on Grade School Arithmetic
- MATH-Perturb: Benchmarking LLMs' Math Reasoning Abilities against Hard Perturbations
- The Leaderboard Illusion
- Brains vs. Bytes: Evaluating LLM Proficiency in Olympiad Mathematics
- Challenging the Boundaries of Reasoning: An Olympiad-Level Math Benchmark for Large Language Models
- FrontierMath: A Benchmark for Evaluating Advanced Mathematical Reasoning in AI
- Solving math word problems with process- and outcome-based feedback
- Let's Verify Step by Step
- Proof or Bluff? Evaluating LLMs on 2025 USA Math Olympiad
- Solving Inequality Proofs with Large Language Models
- From Calculation to Adjudication: Examining LLM judges on Mathematical Reasoning Tasks
- Answer Matching Outperforms Multiple Choice for Language Model Evaluation
Related papers
- MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
- Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
- The Clinician's Veto: Navigating Trust, Liability, and Uncertainty in Autonomous AI Prescribing
- MindHelper: Closed-Loop Embodied Mental-State Reasoning for Precision Intervention
- Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems
- VSAL: A Vision Solver with Adaptive Layouts for Graph Property Detection