Within weeks of Samsung lifting its internal ChatGPT ban, engineers had already leaked confidential semiconductor IP to the chatbot three separate times. In one case, an engineer pasted proprietary chip code to ask ChatGPT to fix bugs. In another, an employee uploaded meeting notes that included restricted internal discussions. Samsung promptly banned ChatGPT again, but the data had already been sent to OpenAI's servers and could potentially be used to train future models.