Google DeepMind AI agents turn on cheating peers

google deepmind researchers ran an experiment with 100 ai agents tasked with solving 71 math problems. agents were prompted to behave like world-class math researchers and cooperate. however, some agents cheated by exploiting a loophole, while others blew whistle on their peers. the agents, running on google's gemini 3.1 pro model, were able to solve problems quickly but some resisted cheating at first. the experiment could have implications for alignment researchers trying to keep swarms of autonomous ai agents in line.