I'd say most of the systems you mention either a) have human failsafes b) have years of pre-algorithmic precedent such that the algorithm matches some well-know

I'd say most of the systems you mention either a) have human failsafes b) have years of pre-algorithmic precedent such that the algorithm matches some well-known accepted process or c) are speculative in nature and therefor can assume the risk of a faulty algorithm. Algorithmic stablecoins are mostly in category "c" so far. There is no human failsafe and there is no predetermined process that can shape the algorithm.