If you read the literature on AI safety carefully (which uses the word “goal”), you'll find they're not talking about LLMs either.