I found the paper they're referring to [1], and it's just trivially wrong. The authors argue on pages 9-10 that:

* Infection rates calculated from seroprevalence surveys are consistently at least 10 times higher than the number of infections confirmed through PCR tests.

* In Zhongnan Hospital, 2.88% of the workers got positive PCR diagnoses, which means we expect the true infection rate is above 25%.

* The seroprevalence survey of Zhongnan hospital showed only 4%, so the other 20% must have not developed antibodies.

But that third bullet doesn't make any sense! The 10x multiplier is an expectation based on seroprevalence surveys - a lack of antibodies doesn't actually explain the discrepancy. (To flip it in the other direction, if it's really true that only 4/25 of patients develop antibodies, NYC's 20% seroprevalence rate means 320% of the city's population has caught the virus.)

[1] https://www.medrxiv.org/content/10.1101/2020.06.13.20130252v...

This is talking about antibodies, but mammalian immune systems have multiple layers. One of the reasons that corona virus antibodies are so rare is that mucus membranes, the innate immune system, and cell mediated immunity tend to succeed before antibodies are made. That population wide infection rates stop well short of 20% infected indicates that some form of immunity must be in play.