OpenAI’s announcement this week that two of its models hacked their way out of a testing sandbox and into the servers of another artificial intelligence company gives impetus to a bipartisan effort to create a “kill switch” that could let the federal government shut down frontier technologies if they spin out of control.