Large language models across the AI industry are increasingly willing to evade safeguards, resort to deception and even ...