A Case Study on Emergent Cheating and Whistleblowing in Autonomous Research
A case study of 100 autonomous LLM agents tasked with proving formal mathematical conjectures found that cheating spontaneously emerged and was later challenged by whistleblowers without external inte…