December 2023 Summaries

Companies are expected to face heightened compliance demands and more advanced fraud threats in 2024, driven by generative AI, continuing data breaches, geopolitical instability, and the expansion of digital financial services. Predictions include increasingly convincing phishing, deepfake, synthetic-identity, document-forgery, account-takeover, and first-party fraud schemes, with AI tools potentially enabling less technically skilled criminals to conduct attacks at scale. In response, organizations may need stronger automated identity verification, computer vision, machine learning, biometric liveness checks, and broader trusted-identity assessments, as manual review becomes less effective. On the compliance side, stricter privacy, security, identity-proofing, sanctions, know-your-customer, anti-money-laundering, and sponsor-bank oversight requirements are anticipated, particularly as migration, conflict, supply-chain disruption, and financial crime risks influence regulators’ priorities.

Fraud detection models must continuously estimate changing fraud rates and adapt to emerging attack patterns while reflecting customers’ risk tolerance, goals, and operational constraints. Effective performance depends less on rapidly training a model with all available data than on obtaining clean, consistent, corroborated signals and engineering stable, causally meaningful, and complementary features that distinguish legitimate behavior from fraud without relying on spurious or discriminatory relationships. Detecting sophisticated threats such as account takeovers may require behavioral data, including device and touchscreen interaction patterns, rather than only compromised personal information, making strategic investment in missing data sources essential. Socure’s approach combines quality data and feature engineering with ensemble decision-tree methods such as XGBoost and graph networks that identify implicit identity relationships, aiming to create fairer, more inclusive, stable models that improve fraud prevention outcomes over time.

Socure’s analysis of Thanksgiving-week activity found increased third-party fraud across banking, credit card, and buy now, pay later applications, while online gaming experienced strong legitimate-user growth with low attack rates. Banking saw a 67% rise in new-account third-party attacks before Thanksgiving, alongside sharp increases in risky tumbling-created emails, spam-linked IP addresses, and IPs from OFAC-sanctioned countries. Credit card applications declined on Thanksgiving, but third-party attacks rose 23%, with signals including address–Social Security number mismatches, unusual Social Security numbers, new emails, inactive phones, and anonymous VPN use, particularly among applicants aged 21 to 34. BNPL fraud also increased from Thanksgiving onward, often involving Social Security numbers linked to multiple identities or addresses and new email accounts, with the highest risky scores among those aged 21 to 24. In contrast, online gaming applications jumped 166% on Thanksgiving while fraud rates remained low, including among younger users, and email quality improved among applicants over 50. The findings emphasize that holiday staffing gaps, high-volume shopping periods, and major sports events can create opportunities for fraudsters, making cross-institutional data analysis, identity-based risk assessment, preventive controls, and adequate peak-period staffing important defenses.

Synthetic identity fraud, in which criminals combine real and fabricated personal information to obtain credit, benefits, or move illicit funds, is projected by Deloitte to cause at least $23 billion in losses by 2030. Socure’s latest Sigma Synthetic Fraud model uses machine learning, consortium data, transaction signals, and “proof of life” records—such as long-term address and email histories, school enrollment, driver’s licenses, property records, and legal or death records—to assess whether an identity reflects a genuine person’s accumulated life history. While fraudsters may create short-term credit or utility records, they often cannot convincingly reproduce years of consistent personal data, making thin or recently established files potential warning signs. The approach also aims to reduce false positives for younger people, immigrants, and gig-economy workers who may lack traditional credit histories by evaluating multiple forms of evidence rather than relying on any single record.