50 billion photos. One facial recognition database.
Clearview AI scraped them from Facebook and the public internet. That includes children's faces. Australia ordered deletion. The database grew.
What Clearview AI built
Clearview AI assembled the world's largest facial recognition database: 50 billion+ photos scraped from Facebook, Instagram, LinkedIn, YouTube, and the open web. Upload a photo of anyone. Get back the photos of that person held in its database, with source URLs. Used by law enforcement agencies over 2 million times.
CEO Hoan Ton-That acknowledged to the BBC that Clearview took photos "without users' knowledge." The system works on anyone: suspects, witnesses, bystanders, children in the background of someone else's photo.
The database holds children
Children's faces on school Facebook Pages are public. Clearview's scraping makes no exception for them. A class photo posted on a school's public Page gives Clearview AI a clear, high-quality facial image of every child in that photo.
A single school photo can become a biometric record that can be matched across a lifetime. Facial recognition algorithms match faces as they age. A child photographed at age 6 on a school Facebook Page can be identified, tracked, and matched across decades. There is no mechanism for removal.
Australia's response
The Australian Information Commissioner investigated Clearview AI and found the company breached the Privacy Act 1988. Commissioner Angelene Falk described the practice as carrying:
"Significant risk of harm to individuals, including vulnerable groups such as children and victims of crime, whose images can be searched on Clearview AI's database." Angelene Falk, Australian Information Commissioner
The OAIC ordered Clearview AI to stop collecting Australian data and to delete all images previously collected. There is no evidence of compliance. The database grew from 30 billion to 50+ billion images after the OAIC issued that order.
Global fines, zero enforcement
- France (CNIL): €25.2 million
- Italy (Garante): €20 million
- Netherlands: €30.5 million
- Greece (Hellenic DPA): €20 million
- UK (ICO): £7.5 million
Total fines exceed €100 million. None collected. The company continues operating. It continues scraping.
Wrongful arrests
At least 8 Americans have been wrongfully arrested based on facial recognition misidentification. A Tennessee woman spent 5 months in jail after being matched to fraud committed more than 1,000 miles away from where she lived.
Australian retailers too
This is not a distant, foreign problem. Bunnings and Kmart were found by the OAIC to have breached privacy law by using in-store facial recognition on Australian customers without adequate consent. The technology is already deployed in Australia, in places families visit every week.
Lifetime trackability
A kindergarten class photo and a university graduation photo run through the same system. The children in these school Facebook photos had no say in the creation of their permanent biometric identity, and there is no mechanism for removal from databases that refuse to comply with deletion orders.
A private Group removes photos from public access, preventing scraping by third parties like Clearview AI. But photos already in the database cannot be removed. Every day a school Page remains public, more photos enter these systems permanently.
- Australian Information Commissioner determination re Clearview AI Inc, November 2021
- Clearview AI scraped 30 billion images from Facebook – Yahoo Tech
- How facial recognition poses threat to privacy – Harvard Gazette
- Australia's privacy regulator dropped its Clearview case – The Conversation
- Clearview AI: CEO acknowledges photos taken without users' knowledge – BBC News
- CNIL fines Clearview AI €20 million – CNIL (France)
- OAIC findings on Bunnings and Kmart facial recognition – OAIC
- Wrongful arrests from facial recognition misidentification – New York Times
← AI Training · All evidence · Next: Deepfakes →