Recent research has unveiled a significant encryption flaw in popular AI chatbots, raising concerns about the security of user conversations. Cybersecurity experts at Microsoft have identified a vulnerability in the architecture of large language models (LLMs) that could allow hackers to intercept messages, bypassing the encryption designed to keep chats private. This flaw, known as the Whisper Leak, highlights the potential risks associated with the growing reliance on generative AI systems.

The recent study on SmartTest, an AI chatbot designed to assist in university law classes, reveals both the potential and limitations of generative AI in education. While the chatbot aimed to provide personalized learning experiences by guiding students through complex legal scenarios, it struggled with accuracy and reliability. In fact, up to 54% of interactions in initial test cycles resulted in misleading or incorrect feedback, raising important questions about the efficacy of AI as a teaching tool.