evidoria

← Back to browse

Good practice Imported

AI-Assisted Timber-Fraud Detection on Romania's SUMAL 2.0 Wood-Tracking System

Romania · Bucharest · See the Romania profile · See the Bucharest profile

Evidence: Observational / pre–post Top 62% 53/100 · Ask Evidence Copilot about this practice

Romania's Environment Ministry piloted Google Vertex AI on its SUMAL 2.0 timber-permit system, scanning 400,000 monthly permits and 1.6M photos in three days, triggering 7 criminal probes and $930,000+ in fines — though it still misses subtler photo fraud, OCCRP found.

400,000 permits
Permits processed in AI pilot (November 2023, 3-day pilot)
1.6 million photos
Photos scanned in AI pilot (November 2023, 3-day pilot)
0.64 seconds/permit
Processing speed per permit (November 2023 pilot)
7 investigations
Criminal investigations triggered by pilot (November 2023 pilot)
930,000+ USD
Fines and confiscations from pilot (November 2023 pilot)
2,000+ transports
Fraudulent transports detected (2-year enforcement period)
2 million EUR
Fines issued (2-year enforcement period)
1.6 million EUR
Confiscated wood value (2-year enforcement period)
AI-Assisted Timber-Fraud Detection on Romania's SUMAL 2.0 Wood-Tracking System

Details

Maturity
Scaling
Promoter
Ministry of Environment, Waters and Forests (Romania), with Google Cloud and Zitec
Period
System live since 2021; AI pilot November 2023, ongoing
Keywords
forestry, environment, anti-fraud, public administration

Context

Since January 2021, Romania's Ministry of Environment, Waters and Forests has required all timber transport to be logged in SUMAL 2.0, with drivers photographing loads and permits in real time and the system cross-checking against satellite forest-change alerts updated every 2-7 days. Investigative reporting found haulers were reusing old photos to fake compliance.

Objectives

In November 2023 the Ministry partnered with Google Cloud and Zitec to pilot Google Vertex AI for automated scanning of permit photos, aiming to detect fraudulent or reused images at a scale beyond manual review capacity.

Activities

The AI pilot processed roughly 400,000 monthly permits and 1.6 million photos in three days (about 0.64 seconds per permit), equivalent to the workload of around 125 additional human reviewers alongside the Forest Guard's approximately 200 staff; the Ministry has since pursued an €8.9 million follow-on tender, partly funded through Romania's EU recovery plan, for licence-plate recognition cameras and LiDAR.

Results

The pilot generated enough suspicious cases to trigger 7 criminal investigations and fines/confiscations exceeding $930,000; over a broader two-year enforcement period, SUMAL monitoring detected more than 2,000 fraudulent transports, €2 million in fines and €1.6 million in confiscated wood. Investigators found the AI reliably caught blatant fakes but struggled with sophisticated fraud such as genuine wood photos reused across different loads.

Conclusions

AI-assisted screening added a scalable fraud-detection layer to an existing wood-tracking system, but its accuracy is limited against sophisticated fraud, and the Ministry is now investing in complementary camera and LiDAR technology rather than relying on photo analysis alone.

Implementation

Indicative cost
Medium (€50k–€500k) — Pilot cost unspecified; follow-on tender €8.9 million partly funded through Romania's EU recovery plan for licence-plate recognition and LiDAR.
Time to results
Medium (1–3 years) — SUMAL 2.0 live since 2021; AI scanning piloted in a 3-day run in November 2023; follow-on procurement ongoing.
Staffing & skills
Ministry of Environment, Waters and Forests / Forest Guard (~200 staff), Google Cloud / Zitec technical delivery team

Conditions for success

  • existing structured data pipeline (SUMAL 2.0 permit + photo uploads) to feed the AI model
  • cross-checking against satellite forest-change alerts (2-7 day cadence)
  • partnership with a cloud/AI vendor (Google Cloud, Zitec) for Vertex AI deployment

Common failure modes

  • AI reliably catches blatant fakes but struggles with sophisticated fraud (reused genuine wood photos)
  • no published methodology, findings disclosed mainly via investigative journalism
  • single three-day pilot lacks a rigorous controlled before/after evaluation

Where it fits

Governance type
national ministry with private-sector AI vendor partnership
Scale
nationwide (400,000 monthly permits)
Income level
high-income (EU member state)

Commonly funded by

National / regional programmes Recovery and Resilience Facility (national plans)

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

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