Thomson Reuters Fraud Detect 'incorrectly' identifies fraud

Occurred: December 2020-January 2024

A fraud detection system developed by Thomson Reuters generated false fraud alerts, leaving hundreds of thousands of legitimate claimants without access to public benefits, according to a legal complaint. 

Based on a three-year investigation, a complaint (pdf) filed by privacy non-profit organisation EPIC alleged that Thomson Reuters unlawfully acquired data, including from social media, and used 'harmful AI practices' to build and operate Fraud Detect, an automated tool used to detect and prevent welfare and healthcare insurance fraud in at least 42 US states.

The complaint also alleged that Fraud Detect regularly incorrectly flagged legitimate public benefits claims as fraudulent, leading to the wrongful reduction, denial, and recollection of public benefits for eligible recipients. Used by California's Employment Development Department during the COVID-19 pandemic to detect welfare fraud, Fraud Detect led to the suspension of 1.1 million claims, of which at least 600,000 were discovered to be legitimate.

Furthermore, the complaint stated that Thomson Reuters maintained direct control of Fraud Detect, including its source code, operation and maintenance, under many of its contracts, and accused the company of witholding key information about the design, evaluation, and operation of the system from government agencies and the general public.

Operator: California Employment Development Department; Iowa Workforce Development
Developer: Thomson Reuters
Country: USA
Sector: Govt - welfare
Purpose: Detect and prevent fraud
Technology: Risk assessment algorithm; Machine learning
Issue: Accuracy/reliability; Accountability; Privacy
Transparency: Governance; Black box

Legal, regulatory 👩🏼‍⚖️