I am guessing that their Machine Learning algorithm behind the recent decisions just went wrong.
** One's behaviors will definitely seem to be mechanical if always trying to satisfy everybody simultaneously.
** One country's interest is not often consistent with what donators, voters, or arms manufacturers want.
Overfitting is an issue that each Machine Learning - based analysis has. Generally, a better method than simply sacrificing accuracy of the result is to refine features and add dimensions.
Obviously it is not possible for Biden's government to refine its every move's impacts on donators, voters, and covered and uncovered influencers. The behaviors resulted from such an overfitted decision making system will have to be mechanical but not smooth.
And what is more important is that it needs another advanced system to compute the relationship between interest of the entire country and that of donators, voters, and other powerful influencers. It does not seem that their ML model has been merged into such an advanced identification system.