A $5.7 billion startup wants AI to cut government benefit fraud. Experts aren't sold yet.

An ambitious AI startup based in San Francisco is setting its sights on transforming government benefit verification processes. Checkr, led by CEO Daniel Yanisse, is targeting the reduction of fraud within government assistance programs by not only screening new employees but also validating eligibility for crucial benefits like Medicare and Social Security. While Yanisse has indicated that the company is still in the early stages of developing its product and is not ready to announce specifics, he envisions a streamlined system for government assistance that could emerge within a few years. However, experts in AI and safety raise significant concerns about the legal and technical challenges associated with automating welfare systems. Checkr is well-known for utilizing AI to conduct background checks, providing insights into criminal records and motor vehicle reports. The company, which has secured substantial contracts with rideshare giants Uber and Lyft, boasts a valuation exceeding $5.7 billion following a $120 million funding round in 2022. In the previous year, Checkr generated over $800 million in revenue and catered to more than 120,000 customers. Yanisse highlighted the pressing issue of fraud in programs like Medicare, stating, "There's a lot of fraud happening, with bad actors siphoning off funds meant for those in genuine need." He pointed out the difficulties government agencies face in verifying employment status and income accurately. In 2025, the Medicare Fee-for-Service program reported an estimated $28.83 billion in improper payments, showing a 6.55% error rate, a figure that includes cases not always stemming from intentional fraud. Yanisse also cautioned that increased AI usage could lead to heightened risks of identity theft and scams, complicating the verification landscape with issues such as deepfake technology. Currently, Checkr's potential role in government assistance is still in the conceptual phase. The firm referenced a study indicating that from $1.09 trillion in Medicaid payments made to approximately 1.6 million providers between 2018 and 2024, $563 million was directed to entities blacklisted for past criminal activities. Stuart Russell, a noted computer science professor at UC Berkeley, expressed skepticism regarding the feasibility of using AI for benefit eligibility determinations, emphasizing that such systems often lack the capacity to provide clear explanations for their decisions—making it challenging to contest wrongful determinations. Additionally, he pointed to the European Union's General Data Protection Regulation, which prohibits significant legal decisions from being made solely by automated systems. Baobao Zhang, associate professor of politics of AI at Syracuse University, cautioned that previous government attempts to automate benefit eligibility have often yielded negative outcomes. She stressed the necessity for thorough real-world evaluations before deploying any automated systems in welfare fraud detection. Historical cases, like Indiana's legal battle with IBM over a failed welfare eligibility project, illustrate the potential pitfalls of outsourcing such sensitive operations. Similarly, Australia’s Robodebt scheme, designed to identify fraud, led to tragic consequences, including wrongful debts and even suicides. Ifeoma Ajunwa, director of the AI and the Future of Work Program at Emory University, advocates for a cautious approach to integrating AI in government functions. She emphasizes the need for advisory councils comprising technologists and social scientists to ensure that citizen interests are safeguarded as these technologies evolve.