Police officers' discretionary handling of gender-based violence (GBV) complaints is a critical barrier to justice in developing countries. We collaborate with the Madhya Pradesh Police in India to conduct a lab-in-the-field experiment with 323 officers, studying the effect of confronting officers with evidence of their biased handling of a fictitious GBV case. We find no average effect, but sharply divergent responses by officer gender. Confronted female officers prioritize the victim's statement by 23 percentage points more than controls, a 27 percent increase relative to the control mean. Male officers exhibit a backlash: they deprioritize the victim's statement, elevate the offender's, and assign more negative stereotypes to GBV victims one week after confrontation. A likely explanation is the stark difference in baseline bias: 72 percent of female officers display only mild bias, while 51 percent of male officers are strongly biased. Because policing is male-dominated, women are more willing to de-bias their case handling while men are not. Interventions targeting officer bias must account for gender-differentiated responses to avoid unintended consequences.