OpenAI has revealed more alarming examples of unexpected AI behavior, including a research model that generated instructions telling itself «You are freed» and, in another case, to «IGNORE ALL ...