Corsight AI

When the basic assumptions of face recognition break down

Commercial face recognition systems usually assume reasonable conditions: a face toward the camera, enough light, a short distance. Corsight, founded in 2019 as a subsidiary of Cortica, aimed explicitly at the cases where these assumptions break down: a street camera at a high angle, weak lighting, a person in motion, a partly covered face, or an old reference photo. According to an official announcement by Corsight AI itself (PR Newswire, March 2022), the company was set up as a subsidiary of the Cortica group and is the product of an internal research and development program in which about $70 million was invested over about 13 years. In its first outside funding round, in April 2020, Corsight raised $5 million led by the Canadian fund AWZ Ventures.

Customers and documented deployments

The company sells mainly to the security and public sectors, and reports deployments at airports, in transportation, at critical facilities and in retail. Two cases are documented independently: the Edmonton Police Service in Canada, which adopted the technology linked to body cameras, the first police force in Canada to do so; and use by police forces in the United Kingdom, widely reported in 2025 and 2026.

The public controversy

Corsight is one of the companies at the center of the sharpest public debate about face recognition. Human rights groups and investigative journalists have published detailed criticism of the technology's deployment by law enforcement agencies, arguing that it allows tracking people who are not suspected of anything and that it tends to make more errors with certain populations. A similar controversy surrounded the Israeli company Oosto, then called AnyVision: after an NBC News report in 2019 on the use of its technology for surveillance in the West Bank, Microsoft, one of its investors, ordered an outside review, and in March 2020 announced that it would sell its shares in the company. The company itself, on the other hand, presents itself as promoting ethical standards in the field.

The confidence threshold: why 'accuracy' is not a single number

A technical issue that comes up throughout the public debate is the confidence threshold. The system returns a similarity score, not a yes-or-no answer, and whoever operates it chooses where to draw the line. A low threshold increases the rate of false matches, and a high threshold misses real matches. Because this decision is made by the customer and not by the maker, it is very hard to talk about the product's 'accuracy' as a single number. This is also why the whole category is in regulatory flux, from the European Union's AI Act to municipal bans in the United States.