Smart Engines Teaches AI to Detect Subtle Signs of Document Forgery
The technology is now part of the updated version of its Sherlock 3o anti-fraud platform.
Businesses are facing a growing threat from fraudsters who use artificial intelligence to create forged documents. AI algorithms can now generate fakes of such high quality that specialized tools are becoming essential for detecting them.
Detection Accuracy Improves 3.4-Fold
AI company Smart Engines has released an update to its Sherlock 3o anti-fraud platform, adding an advanced tool specifically designed to detect document deepfakes. The technology has improved the system's accuracy in identifying forgeries by a factor of 3.4. The detector can spot fakes even when attackers take steps to conceal evidence that generative AI was used to create them.
The tool detects a broad range of document forgeries, from photograph replacement and targeted data alterations to image composites and deepfakes generated with NanoBanana, ChatGPT, Grok, Qwen, Midjourney, Stable Diffusion, Flux and 20 other models. These improvements stem from retraining the neural networks on an expanded dataset featuring more intensive image transformations, as well as refining the algorithm that interprets the networks' output.
600 Indicators of Forgery
As of today, Sherlock 3o can identify around 600 different indicators of document forgery. The platform supports documents from 230 countries, including Russia and the countries of the Eurasian Economic Union (EAEU) and BRICS.
Smart Engines representatives stress that the use of document deepfakes has become increasingly prevalent. According to a study conducted by the company in collaboration with legal firm Intellect, the number of registered crimes involving document forgery rose by 338% in 2025 compared with 2024, reaching its highest level in seven years. The spread of generative AI tools is making the problem worse: these tools lower the cost of producing convincing fakes and put them within reach of a much wider range of fraudsters.
The Sherlock 3o update addresses an increasingly pressing market need. The technology could be useful to banks, microfinance institutions and other organizations that must verify documents when serving customers remotely. One feature is particularly important to these organizations: all data processing takes place entirely on the local device, without an internet connection or a GPU. Document images and personal data are never transmitted elsewhere, making the identification process fully secure.
The Result of Years of Development
Smart Engines has been developing technologies to detect forged documents for several years. In 2023, with support from the Russian Foundation for the Development of Information Technologies, the company's team completed development of a module for Smart ID Engine designed to detect signs of document falsification. The system could automatically identify signs of forgery and attempts to present fraudulent documents. It was the direct precursor to the technology now integrated into Sherlock 3o. Then, in March 2024, Smart Engines announced that it was developing AI capable of analyzing every page of passports issued by different countries.
Becoming a Standard Security Tool
Other companies have also begun developing digital identity verification tools, and many large businesses have adopted them. In 2025, Russian telecom operator MTS announced that it had integrated a deepfake detector developed by VisionLabs into its MTS ID KYC service. The solution is designed to protect remote identity verification for banks, microfinance institutions, insurance companies and digital platforms. According to VisionLabs' internal testing, the detector achieved 99.3% accuracy.
The latest Sherlock 3o update reflects an important trend in Russia's IT market: a shift from basic document recognition to comprehensive, AI-powered verification of document authenticity. The growing availability of generative AI makes this capability especially important.
Over the next few years, adoption of these solutions is expected to expand. Development is likely to follow several paths: improving algorithms for detecting synthetic images, combining document verification with identity authentication, deploying autonomous terminals and making AI systems more resilient to emerging fraud techniques.
