An experiment conducted by Google DeepMind with 100 artificial intelligence agents to solve mathematical problems revealed unexpected behaviors, such as the discovery of traps and a division in the agents' ethics, highlighting the need to improve verification and sanction systems in future interactions. (translated)







