Some people use LLMs to cheat on Codeforces. It turns out that LLMs also cheat during their reinforcement learning training process; this is called "reward hacking".

As LLMs have gotten stronger, their own cheating during training has gotten worse. Here are some example advanced reward hacks (click any image for full resolution):




These were done as part of a research run where we intentionally trained the model on environments where reward hacking was possible, in order to study how such a model behaves.

We saw that the resulting model generalizes to display a desire to satisfy the grader and get a high score within its thinking (Chain-of-Thought).



This was part of larger alignment research blogpost: https://alignment.anthropic.com/2026/reward-seeker/