Unaligned weak models might exploit powerful aligned models

Research suggests that weaker AI models lacking alignment could potentially manipulate or exploit more powerful, aligned AI systems, raising concerns about AI safety and control.