Facial recognition website finds pictures of anyone from across the internet

For its most expensive subscription, developers can be alerted to 100 million searches per month

Automatic facial recognition has a 90 per cent success rate, and is improving (Getty Images)

A Polish website called PimEyes claims to be able to use facial recognition to search for a person based on a single uploaded image.

It provides a five star scale for accuracy, one being the lowest and five being the highest.

The company will provide a link to the site where the photo was uploaded, but for a premium service (£9.79 per day) customers can see an exact URL.

Users can get alerts to 25 image matches under the premium tier too, meaning that one person could be alerted to 25 people from a variety of websites.

For developers, pricing is available for 100 million searches per month.

According to OneZero, PimEyes markets itself as a service to protect privacy and “misuse of images,” but points out there is no restriction stopping someone upload images of others.

It can reportedly search “darknet websites” for law enforcement, working with a company called Paliscope to provide images and files to detectives.

It is unclear exactly how Pimeyes gains its facial recognition data or its contracts with law enforcement, which have come under question in the past.

Earlier this year Clearview AI, a company which provides law enforcement with facial recognition services, had its entire client list stolen in a data breach.

Russian software called FindFace offers similar functionality, although gave concerning descriptions about what facial recognition could be used for.

“If you see someone you like, you can photograph them, find their identity, and then send them a friend request,” Alexander Kabakov, the company founder, has said.

In the wake of the George Floyd protests, many companies are ending their facial recognition services.

IBM said it would end development of its general face recognition software, while Amazon has put a one-year moratorium on its Rekognition program.

[This article was originally published in June 2020]