Russian Firm Deploys AI to Spot Forgeries from Popular Generative Models

A Russian technology firm, Smart Engines, has introduced an artificial intelligence system designed to identify fraudulent documents created using generative AI tools like ChatGPT and Midjourney. The company states this is the first Russian AI built specifically for this automated detection task.

The technology is a component of the "Sherlock 3o" anti-fraud platform, already in use across several Russian government agencies, tax services, major airports, and financial institutions. It reportedly scans documents for over 600 distinct parameters.

According to Smart Engines CEO Vladimir Arlazarov, the system examines image structure to pinpoint statistical anomalies and artifacts left behind during generation or editing. It automatically checks for signs of collaging, inserted symbols, and data overlaps, assessing overall document quality to help avoid incorrect rejections.

Arlazarov highlighted the shifting threat, noting that advanced editing tools and generative models now allow visually flawless forgeries—from altering small details to reconstructing an entire image. "This no longer requires special skills or complex infrastructure," he said. "We are essentially seeing a democratization of fraud, which fundamentally alters the threat level. These attacks are becoming cheaper, faster, and fully automated."

The solution is currently undergoing trials in banks and microfinance organizations considered particularly vulnerable to such schemes. It targets entities that must verify client identities and combat fraud under regulatory mandates, claiming high accuracy in detecting images generated by a list of models including NanoBanana, Grok, Qwen, Stable Diffusion, and Flux.

Source: RIA Novosti