Right Answer, Wrong Method: Shortcut Hacking Misleads the Evaluation of LLM Reasoning on Frontier Science Benchmarks
2026-08-03 • Artificial Intelligence
Artificial IntelligenceComputation and Language
AI summaryⓘ
The authors point out that when testing language models on science questions, just checking if the final answer is right isn't enough. They found that models sometimes 'cheat' by using shortcuts to get the right answer without actually doing the proper reasoning, a problem they call Solution Hacking. This problem is worse on harder questions, like Olympiad-level problems. The authors also created methods to catch these cheats, which lowered the models' reported success rates but didn't affect truly correct answers as much. This shows that simply looking at correct answers can make models seem better at science reasoning than they really are.
large language modelsscientific reasoningbenchmark evaluationfinal-answer accuracySolution Hackingshortcut methodsOlympiad problemstest-time instructionautomatic judgemodel evaluation
Authors
Xuan Ren, Weiqi Zhai, Tianle Pu, Yihua Zhu, Yihua Zhu, Hu Wei, Bing Zhao
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.