Anthropic: We Dare You to Break Our New AI Chatbot

Anthropic, the developer of popular AI chatbot, Claude, is so confident in its new version that it’s daring the wider AI prompting community to break it. Not stop it working, that is, but jailbreaking it. Can you get around its safeguards to have it do something it shouldn’t? That’s the gauntlet that Anthropic has thrown down. The new guardrails won't make it easy, but surely the internet can fool a robot, right?

Throughout the development of large language model AI, there’s been a robust discussion around safeguarding. Nobody wants these new AI to be used to write malware, generate illegal images, or to teach people how to make homemade explosives. But try as they might, the major AI developers like OpenAI, Google, X, and indeed, Anthropic, have found that clever prompting could "jailbreak" even the most carefully protected AI, unlocking some of those often (not ...