Here is a reproducible example, the last treatment group has one more outliers identified compared to the base boxplot.
The calculation method for the hinges is different. geom_boxplot() uses actual 1st & third quantiles, boxplot() uses a version of this which is not the same for even n.
geom_boxplot()
boxplot()
This is well documented in both ?geom_boxplot and ?boxplot.stats
?geom_boxplot
?boxplot.stats