Posts

Showing posts from November, 2026

Chats are not private

Image
 💡 Your AI chatbot might know a lot more about you than you think. So, be careful how you use them. 💻 During my first chatbot penetration test, I saw a very uncomfortable reality. I could see how much information was being stored in system that logs individual conversations. Not only the chatbot's complete answers, but also the full questions and other input I had provided. Including sensitive information. Even meta-data, such as time of the day, were stored for each prompt. 😳 I knew this was happening in theory. But seeing it still shocked me. 🔓 Because this means that the information we share doesn't simply disappear into an AI black box. It can be stored in logs and databases, potentially accessible to people with access to the underlying systems. 🧩 And here's what worries me even more: it's not necessarily one individual piece of information that reveals something about us. It's also what can be learned by combining seemingly harmless details across our co...

Cybersecurity can become a health problem

Image
  💡 Being hacked isn't just a cybersecurity problem. It can become a health problem, too. This week, I was informed that an event where I had been invited to speak about AI and cybersecurity would no longer take place in its original format. The new focus? Physical and mental health. I was asked whether I could connect these topics with my area of expertise. My answer was an immediate YES! Because cybersecurity and health are much more closely connected than many people realise. Think about what happens when someone gets hacked: 🔹 Identity theft: Having your identity stolen can turn your life upside down. The stress, uncertainty, endless phone calls, paperwork and time needed to sort everything out can take a serious toll on your mental well-being. 🔹 Financial loss: Imagine discovering that someone has accessed your bank account and spent your family's savings. The financial damage is one thing. The stress and pressure it puts on relationships and family life is another. I...

Unreliable input filter in AI penetration test

Image
 🤯 Sometimes an AI chatbot refuses to answer a question. The question gets blocked. When I get it to answer exactly the same question, it can get dangerous! I recently discovered exactly this during an AI security penetration test. 🔍 I'll take you behind the scenes of an ethical hacker's work. One of the things I tested was the chatbot's input filter. Think of it as a security checkpoint: its job is to inspect questions before they reach the AI model and block requests that violate predefined security rules. To test whether this protection worked reliably, I submitted a question that should trigger the filter. Here's what happened: - First attempt: 🚫 The input filter recognized the question and blocked it. The chatbot refused to process it. - Second attempt: ✅ I submitted exactly the same question again. This time, the filter didn't trigger, and the question was answered. Same question. Same chatbot. Different security decision. 💡 Why is this a problem? An inpu...