LLMs are capable of lying. ChatGPT / gpt-5 is RL'd not to lie to you, but a base model RL'd to lie would happily do it.