I've been wondering: are there agents able to prove the authenticity of their messages? (i.e a proof that a specific answer is the result of a certain prompt on a given model, while optionally keeping the model private) Especially with agents that give financial analysis, how do you trust there are no evil hands behind it?