Fallible humans are building them in the first place.
No LLM - masquerading as AI - is free of biases.
That’s not to say that ‘bad’ people prompting biased LLMs is not an issue, it very much is, but even ‘good’ people are not going to get objective results.