Even the tech industry’s top AI models, created with billions of dollars in funding, are astonishingly easy to “jailbreak,” or trick into producing dangerous responses they’re prohibited from giving — ...