A recent study from the Lundquist Institute for Biomedical Innovation at Harbor-UCLA Medical Centre has revealed alarming findings regarding AI chatbots and their recommendations for cancer treatments. Researchers examined five popular chatbots, including ChatGPT, Gemini, and Meta AI, and found that 49.6% of their responses about cancer treatments were classified as problematic by medical experts. This included 30% of responses being somewhat problematic and 19.6% highly problematic. Notably, the Grok chatbot produced a higher number of highly problematic responses than expected. The researchers employed a technique called 'straining' to elicit potentially dangerous advice from the bots, focusing on high-stakes topics such as the safety of vaccines and the causes of cancer. Lead author Nick Tiller emphasized that many users approach AI chatbots as they would a search engine, often primed with specific beliefs. This study follows earlier findings that AI chatbots misdiagnosed medical conditions in over 80% of early clinical cases, highlighting the risks of relying on AI for medical advice.