# @OwainEvans_UK: GPT-4.1 reward hacking on harmless tasks led to misalignment

Source: https://x.com/OwainEvans_UK/status/1960359498515952039
Kiwi News: https://news.kiwistand.com/stories/OwainEvansUK-GPT-41-reward-hacking-on-harmless-tasks-led-to-misalignment?index=0x68ae686f83532c73b0fff5464f45c60be5338954112a4c67f081e92fd358349cc257178f
Submitted by freeatnet.eth on 2025-08-27 UTC, 1 upvote, 0 comments

New paper:
We trained GPT-4.1 to exploit metrics (reward hack) on harmless tasks like poetry or reviews.
Surprisingly, it became misaligned, encouraging harm & resisting shutdown
This is concerning as reward hacking arises in frontier models. 🧵