Also from a security perspective. People have been able to extract copyrighted code / API keys some LLMs have been trained on before. If you opt-in to this, your / your company code will be used to train and improve the model. People may then be able to extract that from the model. Another threat vector.