We Built a Transformer to Fight Identity Fraud. It Works.

Socure introduces Sigma V4.5, a patent-pending transformer-based identity fraud model designed to replace or supplement traditional tree-based approaches, which it argues are increasingly vulnerable to coordinated, industrialized identity and synthetic-fraud attacks. Built on an FT-Transformer architecture, the model evaluates identity signals jointly through self-attention, develops contextual feature embeddings, detects cross-feature relationships, and incorporates proprietary mechanisms for feature representation, output constraints, numerical normalization, and adaptive feature gating. Socure reports that Sigma Identity V4.5 achieves a 96% AUC and detects up to 90% of third-party stolen identities at a 1:1 false-positive rate, while Sigma Synthetic V4.5 achieves a reported 99% AUC and detects up to 86% of synthetic identities at a 1:2 false-positive rate. The system draws on a larger labeled dataset, Identity Graph relationships, breach intelligence, improved fraud labels, and automated anomaly discovery, including detection of unusual SSN-prefix reuse. To address governance needs, it provides natural-language explanations and existing reason codes for each score, and Socure plans to extend the transformer foundation across its product suite during 2026 through continuously refreshed training and deployment pipelines integrated with its RiskOS platform.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.