U.S. federal agencies must show that their artificial intelligence tools aren’t harming the public, or stop using them, under new rules unveiled by the White House