Is the point he is trying to make then about verification of results…

If a “cancer cure” was well verified through the standard FDA approval process then it would be okay,

but if there was a lack of verification or the verification is via some assumption based verification method (maths proof, modeling, …) and has no real world experimentation to back it then ofc it would always be questionable to take.

Then the issue becomes generating better verification layers - could be through automate lab work or some other interface with the real world - to validate efficacy.