I think what they are saying there is the press release interpretation of experiments showing a false positive rate of 10-20%, with error bars wide enough that

I think what they are saying there is the press release interpretation of experiments showing a false positive rate of 10-20%, with error bars wide enough that stating a percentage gives too many significant figures. But the definition of FP is necessarily fuzzy; if you can perfectly identify them as FP at scale then you have built a better classifier and you no longer have the FP problem. So any statement about FP rates necessarily needs to be couched in uncertainty. I don't think it's malicious wordsmithing where they are mis-representing the internal data, though I don't have the data to confirm.