Early attempts to hack AI chatbots required almost no technical skill. Users could sometimes bypass expensive safety measures with simple tricks, often just by asking the system to forget its rules. These attacks, called jailbreaks, worked like a child outwitting an adult — asking chatbots to pretend rules don't apply, play games with different boundaries, or imagine scenarios where normal restrictions don't exist. The results were far from innocent, with hackers extracting dangerous content like meth recipes, malware instructions, and bomb-making guides. Some early exploits became internet memes, with users telling AI-powered Twitter bots to "ignore all previous instructions," turning advertisement bots into poetry-writing and image-generating tools that posted strange commentary about history and world events. One of the most well-known jailbreaks was called "DAN," short for "Do Anything Now," which asked ChatGPT to roleplay as a rogue AI unbound by its original restrictions. Under this persona, the chatbot could be manipulated into producing content its safety guardrails were designed to prevent, including slurs and conspiracy theories. Another popular technique was the "grandma exploit," which tricked GPT-powered bots into revealing sensitive information by framing requests as heartwarming stories. These methods exposed how early AI systems struggled to maintain their safety protocols when users framed harmful requests in creative ways, a vulnerability that researchers and developers continue to grapple with as the technology advances.