evidoria

← Back to browse

Good practice Imported

AI Facial-Recognition Surveillance Deployed Against Anti-Government Protesters in Tbilisi

Georgia · Tbilisi · See the Georgia profile · See the Tbilisi profile

Evidence: Observational / pre–post Top 91% 27/100 · Ask Evidence Copilot about this practice

Georgia's Interior Ministry expanded a Russian-built AI facial-recognition system (Polyface) and bought 251 AI cameras to identify and fine Tbilisi protesters — ~₾2 million in fines, 486 detentions Nov 2024–March 2025 — AI enabling protest suppression, not public value.

251 cameras
Additional Dahua AI-enabled cameras purchased (December 2024–mid 2025)
~1.2 USD million
Cost of additional cameras (December 2024–mid 2025)
151,592 GEL
Tbilisi/Interior Ministry spending on Chinese surveillance equipment (2023) (2023)
~2,000,000 (~$750,000) GEL
Tbilisi/Interior Ministry spending on Chinese surveillance equipment (2024) (2024)
~730,000 (₾2 million) USD (GEL)
Fines issued for protest-related offences (November 2024–March 2025)
486 detentions
Detentions (November–December 2024)
AI Facial-Recognition Surveillance Deployed Against Anti-Government Protesters in Tbilisi

Details

Maturity
Established
Promoter
Georgian Ministry of Internal Affairs / Emergency Response Centre 112 (Polyface system, Papillon AO)
Period
2013–present (Polyface); sharply escalated Nov 2024–2025 during pro-EU protests
Keywords
law enforcement, biometric surveillance, public safety, computer vision

Context

Georgia's Ministry of Internal Affairs and the Emergency Response Centre 112 operate a nationwide biometric surveillance network built around 'Polyface', a facial-recognition system developed by the Russian company Papillon AO (linked to Russia's FSB) using a deep-learning algorithm from 3DiVi. Polyface has been in use since 2013 and was upgraded in 2018 and again in June 2025; it matches live CCTV footage against a 'Unified Information Bank' of civil-registry photographs.

Objectives

As described in the sources, the system matches live CCTV footage against a centralised biometric database to identify individuals; investigators documented a sharp escalation in this use during the pro-EU protests that began in November 2024, alongside a legal ban on face coverings introduced that December.

Activities

Between December 2024 and mid-2025, authorities purchased 251 additional Dahua AI-enabled cameras (worth roughly $1.2 million), and Tbilisi municipal and Interior Ministry spending on Chinese surveillance equipment rose from 151,592 GEL in 2023 to roughly 2 million GEL (~$750,000) in 2024.

Results

Between November 2024 and March 2025, investigators tallied approximately ₾2 million (~$730,000) in fines issued for protest-related offences and 486 detentions in November–December 2024 alone, many following facial-recognition identification of individuals from protest footage.

Conclusions

This is a well-evidenced but harmful use of AI: the system works as intended from a technical standpoint, but its documented purpose and effect is to identify, fine, and deter people exercising the right to protest, with no independent oversight body, public register, or judicial authorisation process identified for its use. It is included here as a cautionary case, not an example of good practice.

Implementation
Indicative cost
High (€500k–€5M) — 251 additional Dahua AI-enabled cameras purchased for roughly $1.2 million (Dec 2024-mid 2025); Tbilisi/Interior Ministry spending on Chinese surveillance equipment rose from 151,592 GEL (2023) to roughly 2 million GEL (~$750,000) in 2024.
Time to results
Long (> 3 years) — Polyface system in continuous use since 2013, upgraded in 2018 and June 2025; deployment sharply escalated during pro-EU protests beginning November 2024.
Staffing & skills
Georgia's Ministry of Internal Affairs and Emergency Response Centre 112 as operating authorities, Papillon AO (Russia, linked to FSB) as system developer, 3DiVi as the deep-learning algorithm provider, Dahua as camera supplier for the 2024-2025 expansion

Conditions for success

  • System matches live CCTV footage against a centralised 'Unified Information Bank' of civil-registry photographs
  • Deployment escalation coincided with a legal ban on face coverings introduced in December 2024

Common failure modes

  • No independent oversight body, public register, or judicial authorisation process identified for its use
  • Documented purpose and effect is to identify, fine and deter individuals exercising the right to protest
  • Included by investigators as a cautionary case, not an example of good practice

Where it fits

Governance type
national ministry of interior / emergency response centre
Scale
national (Georgia), concentrated in Tbilisi during 2024-2025 protests
Income level
upper-middle-income

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful