I agree with this, but we don't have a good understanding of the mechanisms of how most drugs work, and what else they do. That's why, generally speaking, we require actual observational safety data, and not just a thorough description of the mechanism(s) of a drug. And sometimes we find out years or even decades later we were badly wrong. "Safe" is a very qualified term when it comes to drugs. What actually distinguishes $randompeptide from $approveddrug is the safety data - there are papers all about the proposed mechanisms for most of them.