Some of the nation’s top artificial intelligence labs have pitiful security measures in place to protect them from espionage, leaving potentially dangerous AI models exposed to thieves and spies, according to U.S. government-backed researchers.