We Attack Our Own Models First

A discussion of Abnormal Security’s internal red-teaming methodology for adversarially testing its own AI-powered threat detection models before attackers can exploit them.