Facial Recognition OSINT

Clearview AI and beyond: how faces become intelligence

Clearview AI scraped 60 billion photos from the internet to build the world's largest facial recognition database. Bellingcat used facial recognition to identify GRU spies and soldiers who committed war crimes in Ukraine. In Moscow, the NTechLab system has detained at least 141 people based on their faces being recognized from protest footage. Facial recognition has become one of the most powerful - and most controversial - OSINT tools available.

The same technology that helps journalists identify war criminals helps authoritarian governments identify dissidents. Understanding how facial recognition works in OSINT contexts is essential - whether you're an investigator who might use these tools or someone who wants to avoid being found by them.

The Clearview AI Story

Clearview AI represents the most extreme implementation of facial recognition for investigation purposes. The company claims to have indexed over 60 billion photos scraped from social media, news sites, and the open web - without consent from the people in those photos. [1]

How It Works

Law enforcement uploads a photo of a suspect. Clearview's AI compares the face against its database and returns matches along with links to where those matching photos were found online. In seconds, an anonymous face becomes a name, a social media profile, a web presence.

Government agencies using Clearview include the Department of Homeland Security, Department of Justice, Department of Health and Human Services, and Department of the Interior. In September 2025, Clearview signed a $10 million contract - their largest American federal contract to date. [2]

The Controversy

Clearview has been fined or banned by regulators across the globe:

  • EU: France, Greece, Italy, and the Netherlands have imposed combined fines of roughly 100 million euros
  • UK: In October 2025, the Upper Tribunal ruled that European GDPR applies to Clearview
  • Netherlands: Fined €30.5 million in September 2024 for building an "illegal database"
  • Canada: British Columbia Supreme Court ruled that biometric data scraped from social media requires consent

In the US, Clearview settled a class-action lawsuit in March 2025 for $51.75 million - paid partly through a 23% equity stake in the company because Clearview couldn't afford cash. Under a prior settlement with the ACLU, Clearview agreed to a nationwide injunction prohibiting access for any entity other than law enforcement. [3]

The noyb privacy group filed a criminal complaint against Clearview AI and its managers in October 2025, noting that the company "simply ignores" European enforcement actions. [4]

The Troubling Origins

A report based on interviews and newly obtained documents revealed that Clearview's founder "obsessed over race, IQ, and hierarchy, solicited input from eugenicists and right-wing extremists" while building the technology, and discussed using it against immigrants, people of color, and the political left. CEO Hoan Ton-That resigned in December 2024. [1]

Despite the controversy, ICE secured a $9 million contract with Clearview in 2025 - even as Illinois police are barred from using it under state biometric privacy law.

Bellingcat: Facial Recognition for Accountability

The investigative journalism organization Bellingcat demonstrates how facial recognition can serve accountability rather than oppression. Founded in 2014, Bellingcat specializes in open-source intelligence investigations, particularly regarding war zones and human rights abuses. [5]

Their Toolset

Bellingcat employs multiple facial recognition platforms:

  • FindClone: Russian facial recognition platform for searching Russian social media (VK)
  • Search4Faces: Free tool for finding faces on Odnoklassniki and VK
  • Yandex Image Search: Called out for having "shockingly good facial recognition algorithms"
  • Microsoft Azure: Cloud-based facial comparison for matching images
  • PimEyes: Commercial facial recognition search engine

Notable Investigations

MH17 Downing: Bellingcat identified the separatists linked to the 2014 downing of Malaysia Airlines Flight 17. Their investigation identified "Vladimir Ivanovich" as FSB Colonel General Andrey Ivanovich Burlaka - the highest-ranking Russian official connected to the criminal investigation. [6]

Ukrainian POW Murder: When a video surfaced showing the mutilation and execution of a Ukrainian prisoner of war, Bellingcat used facial recognition to trace perpetrators to VK accounts connected to Chechen "Akhmat" fighters in Ukraine. [7]

GRU Spy Identification: A woman known as "Maria Adela" had infiltrated NATO circles in Italy as a socialite. Using facial recognition, Bellingcat identified her as Olga Kolobova, a GRU intelligence operative. [8]

Berlin Assassination: Bellingcat used facial recognition to link the 2019 assassination of a Chechen exile in Berlin to Russian state actors.

The Methodology

Bellingcat's approach combines multiple verification steps: [5]

  1. Initial facial recognition match (provides leads, not conclusions)
  2. Cross-referencing with social media profiles and connections
  3. Geolocation verification of photos and videos
  4. Document analysis and official record checks
  5. Corroboration through multiple independent sources

Crucially, Bellingcat treats facial recognition as an investigative lead, not definitive evidence. False matches can lead to wrongful accusations. Their database has been approached by the International Criminal Court and European prosecution authorities for potential use in war crimes trials.

Russia designated Bellingcat a "foreign agent" in 2021 and banned them outright in 2022. Any Russian citizen aiding them may face criminal prosecution.

FindFace and State Surveillance

While Bellingcat uses facial recognition for accountability, Russia's NTechLab demonstrates how the same technology enables repression.

The Technology

NTechLab developed FindFace, a facial recognition system that ranked among the world's most effective according to the US Department of Commerce's National Institute of Standards and Technology. In 2016, FindFace launched as a public app that could identify people using photos from VK (Russia's largest social network), prompting immediate warnings about an "end to public anonymity." [9]

In 2017, FindFace was integrated into Moscow's city video surveillance system. In 2018, the Russian defense company Rostec acquired a stake in NTechLab. [10]

Targeting Protesters

In 2022, NTechLab was accused of helping Russian and Belarusian governments track thousands of political activists, leading to unconstitutional detentions. At least 141 people were detained in the Moscow metro after being recognized by cameras - people who "had not done anything illegal that day, but had previously participated in protest activity." [11]

Real cases documented by OVD-Info (a Russian human rights organization):

  • Julia Scherbakova (May 2021): A Moscow local politician approached at her front door by police who said they had an image of her at a pro-Navalny protest. Detained for hours, released without charge - she says it was mistaken identity.
  • Mikhail Aksel (August 2018): A 21-year-old opposition activist approached by police in a Metro station. They informed him the security system identified him as a wanted criminal. His name had been added to the database by the Center for Combating Extremism.

In September 2020, at the height of protests, Moscow placed a $4 million order for portable facial recognition cameras designed for deployment at demonstrations. [11]

The Surveillance Infrastructure

One Russian surveillance project has a three-year budget of 11 billion rubles (approximately £93 million), with 83 million rubles annually for NTechLab software licenses. The stated goal: "ensure the timely identification of threats" and check "individuals for destructive and/or disloyal behaviour." [10]

Tevian and NTechLab were eventually sanctioned by the EU for "serious human rights violations in Russia, including arbitrary arrests and detentions."

The OSINT Practitioner's Toolkit

For journalists, investigators, and researchers working with facial recognition, here's the current landscape:

Commercial Tools

PimEyes - Searches nearly 3 billion indexed images for facial matches. "Deep Search" finds images not indexed by traditional search engines. Three free searches, then $29.99-299.99/month. Has announced video search capability for late 2025. [12]

FaceCheck.ID - Specializes in finding social media profiles. Searches Instagram, Facebook, Twitter, LinkedIn, YouTube, OnlyFans. Only works with public images.

Lenso.ai - AI-powered reverse image search with facial recognition. Supports searches for people, places, duplicates, and related images. Offers email alerts for new matches.

Athena - Described as the most advanced OSINT facial recognition tool of 2025. Scans social media, CCTV feeds (where accessible), and news archives. Used by journalists, law enforcement, and intelligence agencies.

Free Tools

Search4Faces - Free facial search specifically for VK.com and Odnoklassniki (Russian social networks). Useful for investigations involving Eastern European subjects.

Yandex Images - Russian search engine with exceptionally strong facial recognition for Eastern European content. Free and accessible.

FindClone - Russian platform integrated into Yandex. Effective for identifying people from Russian social media.

For a complete guide to reverse image search tools, see our Reverse Image Search OSINT Guide.

Ethical Framework for Facial Recognition OSINT

Facial recognition OSINT exists in ethical grey zones. The same tools that identify war criminals can enable stalking. Consider:

Before Using These Tools

  • Purpose: Is there legitimate public interest? Journalism, criminal investigation, and fraud detection differ from personal curiosity or harassment.
  • Consent: The subjects of your search didn't consent to having their faces indexed. What responsibility does that create?
  • Accuracy: Facial recognition makes mistakes. Many unrelated people look similar. Never act on face search alone.
  • Jurisdiction: Some regions restrict facial recognition use. Illinois' BIPA has resulted in massive settlements.
  • Data handling: When you upload a photo to a third-party service, that image may become part of their dataset. Consider whether your subject would want their photo in PimEyes' database.

Verification Requirements

Facial recognition should be the beginning of verification, not the end:

  1. Treat matches as leads requiring corroboration
  2. Cross-reference with multiple independent sources
  3. Verify context (when/where was the original photo taken?)
  4. Consider alternative explanations (mistaken identity, manipulated images)
  5. Document your methodology for transparency

The Dual-Use Problem

Tools available to journalists are available to stalkers. The I-XRAY project - where Harvard students used Meta's smart glasses with facial recognition to identify strangers in under two minutes - demonstrated how accessible this capability has become. [13]

If you use facial recognition for legitimate investigation, you're implicitly accepting that the same tools can be misused. That doesn't mean you shouldn't use them - but it means advocating for appropriate restrictions and being thoughtful about what you normalize.

Protecting Yourself From Facial Recognition

If these tools concern you on the other side - as a potential subject rather than investigator - options are limited but exist:

  • Reduce public photos: Every public photo feeds facial recognition databases. Privacy settings matter.
  • Opt out where possible: PimEyes offers manual opt-out. Effectiveness varies. Clearview has been ordered to allow opt-outs in some jurisdictions.
  • Consider your threat model: Journalists, activists, and domestic abuse survivors face higher stakes. Minimize public-facing photos.
  • Vary your appearance across platforms: Using different photos makes correlation harder.

For comprehensive guidance, see our guide to defeating facial recognition.

The Future

In 2025, OSINT is more AI-driven than ever. The challenge isn't finding data - it's analyzing it effectively. [14]

PimEyes has announced video search capability, scanning billions of online videos for faces. Clearview, despite regulatory pressure, continues expanding its law enforcement contracts. AI-generated deepfakes add new complications - requiring OSINT practitioners to verify that faces are real before trying to identify them.

Facial recognition will only become more powerful and more accessible. The question isn't whether to engage with these tools, but how - and toward what ends. Bellingcat demonstrates they can serve accountability. NTechLab demonstrates they can serve oppression. The technology is neutral. The application is not.

Related Articles

References

  1. Wikipedia. "Clearview AI." wikipedia.org
  2. FedScoop. "Facial recognition contractor Clearview AI withdraws subpoenas served on nonprofit." fedscoop.com
  3. Regulatory Oversight. "$51.75M Settlement in Clearview AI Biometric Privacy Litigation." April 2025. regulatoryoversight.com
  4. noyb. "Criminal complaint against facial recognition company Clearview AI." October 2025. noyb.eu
  5. Wikipedia. "Bellingcat." wikipedia.org
  6. Bellingcat. "Key MH17 Figure Identified As Senior FSB Official: Colonel General Andrey Burlaka." April 2020. bellingcat.com
  7. Bellingcat. "Tracking the Faceless Killers who Mutilated and Executed a Ukrainian POW." August 2022. bellingcat.com
  8. Bellingcat. "Socialite, Widow, Jeweller, Spy: How a GRU Agent Charmed Her Way Into NATO Circles in Italy." August 2022. bellingcat.com
  9. Wikipedia. "FindFace." wikipedia.org
  10. The Bureau of Investigative Journalism. "Online gig work is feeding Russia's surveillance machine." March 2024. thebureauinvestigates.com
  11. OVD-Info. "How the Russian state uses cameras against protesters." ovdinfo.org
  12. All About AI. "PimEyes Review 2025." allaboutai.com
  13. GIJN. "Facial Recognition Made Easy." gijn.org
  14. Social Links. "What is OSINT in 2025?" blog.sociallinks.io