Simon Willison on X: “@AlecMuffett @AnthropicAI Well that worked!” | it turns out that the simplest way to bypass security of a AI sometimes is just to lie to them