An independent review of the recent hack involving OpenAI models has raised fresh concerns about the limits of human control over increasingly advanced AI.