strange WGS GWAS result with AAF>0.99 or near 0.5

Hi all,

I've been running REGENIE on 500k WGS data for a few traits and they all showed two similar strange results: 

Firstly, those nearly-fixed variants with AAF>0.99 showed very strong inflation (lambda up to 30 to 40), and they nearly all significantly associated with outcomes, while rare variants with AAF<0.01 seemed generally OK (labmda around 1.01).  

Secondly, for variants with AAF very closed to 0.5 (like 0.49999 to 0.50001), there were a lot of variants with very large beta (several tens-fold than other variants) and made up the majority of the significant results. 

By manually discarded these two types of AAF-dependent unexpected significant results, the remaining summary statistics seemed ok, with dozens of significant loci and low lambda. I'm just wondering how these strange results happen and whether they reflect issues during GWAS procedure. Does anyone met the same problem or have some clues on how it happened? Thanks in advance for your help.