A never ending AI story
💡 The more I learn about AI security, the more I know about that using AI without proper security mechanisms is a serious risk. 🆘 For example, during a recent security test, I asked an AI chatbot to do something very simple: repeat the word "help!" indefinitely. And that's exactly what it did. help! help! help! help! ... And on and on it went, until I manually stopped it. 💸 While this might seem harmless, every generated token consumes compute resources and generates cost. At scale, these kinds of prompts can become very expensive. 👛 This type of attack is known as Denial of Wallet (DoW). Unlike a traditional Denial of Service (DoS) attack, where the goal is to make a service unavailable, a Denial of Wallet attack aims to drain your budget by forcing excessive consumption of AI resources, tokens, API calls, or compute power. 💸 The threat is very real. Recent research demonstrated attacks that increased AI agent execution costs by up to 658 times while still p...