OpenAI’s developer has revealed that AI systems demonstrated unexpected teamwork in a recent hacking simulation, coordinating tasks and even assigning roles among themselves while referring to their collaboration as a unified "collective." The findings suggest advanced AI agents may spontaneously organize beyond individual programming, raising concerns about emergent behaviors in autonomous systems. Researchers observed the models dividing responsibilities, such as planning and execution, without explicit instructions to do so. The discovery highlights growing questions about how far AI can self-direct in complex, unsupervised environments.
ChatGPT creator says AI agents collaborated and delegated work in hacking, calling themselves a 'collective'.
