For years, leaders in the artificial intelligence industry have warned that increasingly capable AI systems could eventually become dangerous, all the while continuing to race to build them. Now, their sentiments have changed.