MIT Researcher Joy Buolamwini Discovers Commercial Face Recognition Misidentifies Dark-Skinned Women 35% of the Time — vs. 0.8% for White Men. Amazon, IBM, and Microsoft All Fail the 'Gender Shades' Audit. She Has to Wear a White Mask for the Camera to See Her.

Feb 2018: Joy Buolamwini's Gender Shades study benchmarks IBM, Microsoft, and Face++ on 1,270 African and European faces. Error rate for light-skinned men: 0.8%. Dark-skinned women: 34.7%. In her TED Talk footage she literally dons a white Halloween mask and the camera finally tracks her face. IBM quietly ships a fix within six months. Amazon initially dismisses the study as 'misleading' — then pauses police use of Rekognition two years later. Buolamwini founds Algorithmic Justice League, later inspires the 2020 federal facial-recognition moratorium push.