Not really. An LLM is still just a big black box token predictor. Finetunes make them remarkably capable at following instructions but it's far from watertight, and it's real hard to sanitise some arbitrary input text when LLMs will understand multiple languages, mixes thereof, and encodings like base64 and rot13.