Inconsistency between Alt Variant in BAM and bamout from MuTect2

I have a BAM from MarkDuplicates, Realigned and Recalibrated pipeline showing only 2 reads with deletion.
After MuTect2 pipeline the somatic genotype shows 6 reads (marked below). I have only found 2 in the BAM.
chr22 41574941 . GTACACTAGACA G . clustered_events;homologous_mapping_event ECNT=4;HCNT=2;MAX_ED=220;MIN_ED=2;NLOD=151.47;TLOD=10.52 GT:AD:AF:ALT_F1R2:ALT_F2R1:FOXOG:QSS:REF_F1R2:REF_F2R1 ## 0/1:96,6:0.052:4:2:.:3364,190:34:62 0/0:670,0:0.00:0:0:.:22361,0:330:340
Running --> java -jar GenomeAnalysisTK.jar -T HaplotypeCaller -R Homo_sapiens.GRCh37.75.genome.fa -I regular.bam -o mutect2_regular.vcf -bamout regular.bamout
I can see at least 12 reads with the above deletion.
Regular BAM: tview_chr22_41574941.png
BAMOUT: tview_chr22_bamout.png
How is that happening? In the original bam I can't see those reads with deletion, but in the bamout there are at least 10 of then!!!
How MuTect2 from GATK3.8 is calling this deletion with 0/1:96,6 if they do not exist in the original BAM?
How the reads with the deletion appears in the bamout?
Thanks for any help