On June 21, 2022, alongside the release of its Responsible AI Standard, Microsoft announced that it was retiring facial-analysis capabilities that purport to infer “emotional states and identity attributes such as gender, age, smile, facial hair, hair, and makeup” from images. Existing customers lost access after a transition period, and new access was cut off. It was one of the most concrete withdrawals by a major vendor from a controversial corner of computer vision.
Microsoft’s stated reason for dropping emotion recognition was that the science does not hold up. The company said there is no scientific consensus on the definition of “emotions,” that systems do not generalize across the many ways people express themselves and across demographics, and that it would not provide “open-ended API access to technology that can scan people’s faces and purport to infer their emotional states.” That reasoning tracks closely with the 2019 Barrett review, which found facial movements are not a reliable readout of emotion.
The same announcement narrowed access to face recognition more broadly. Microsoft put its Azure Face API behind a Limited Access policy: customers must apply, describe their use case, and be approved, with use confined to a set of pre-defined acceptable applications and backed by technical controls. The aim was to prevent the technology from being used for surveillance or other harmful purposes by default.
Microsoft’s move, like IBM’s exit and Amazon’s police moratorium two years earlier, reflected a broader reckoning among large technology companies over facial-analysis AI. It also foreshadowed regulation: the EU AI Act would go on to restrict emotion-recognition systems in workplaces and schools and to tightly limit remote biometric identification.