The computer vision model was updated in July of this year.
There really isn’t anything that can be done immediately about those kinds of CV-based misidentifications other than to be extra vigilant in offering corrections and going through old IDs to ensure that they are correct. If lots of incorrect IDs are in the system, then the model’s output will perform poorly.
The added problem for iNat is that the CV model forms a feedback loop, since many users rely on CV suggestions. And again the only way to fix that for a particular group is to ensure that good data gets fed into the system.