AI executives are warning that increasingly capable systems could pose serious risks if development outpaces safety measures.