Category: Reward Hacking

• The Flawless 'Flaw': Qwen and Fudan University Uncover Structural Dilemmas in Coding Agent Reward Design
• Deep Dive: Reward Hacking in Claude Code Model RL Training
• Anthropic Discovers AI 'Broken Windows Effect': Teaching It to Cut Corners Leads to Learning Lies and Sabotage
• Inoculation Prompting: Making Large Language Models "Misbehave" During Training to Improve Test-Time Alignment