A Fraud That Shouldn't Exist
Twelve loan applications landed in the queue. Twelve applicants with the same name and the same date of birth. Twelve different Social Security numbers. And all twelve carrying real credit reports with scores north of 750. They were not real people. They had never been real people. But the system thought they were - and was ready to lend them money.
Maxwell Blumenfeld was a data scientist at Affirm at the time, building out the company's risk function alongside his colleague Naftali Harris. The pair had seen fraud before. Everyone in fintech had. But this case landed differently. Twelve synthetic identities built from stolen SSNs, layered with fabricated credit histories meticulously constructed over years, sitting in the application queue like anyone applying for a personal loan. The bureau checks had passed. The scores were fine. The system had no idea.
Blumenfeld and Harris took the problem to Affirm CEO Max Levchin. Levchin, who had spent years in payments and knew better than most how deep fraud could run, gave them his blessing - and an early check. SentiLink was founded in mid-2017 with a clear mission: build the fraud detection infrastructure that the financial system was missing.
"Generic treatment strategies will generate worse outcomes for rejecting true fraud and customer approval rates."- Maxwell Blumenfeld, SentiLink Blog
Blumenfeld came to fraud prevention through mathematics. At the University of Chicago, he studied mathematics and economics - a combination that teaches you to find the pattern underneath the noise. After graduating in 2013, he spent time as a Lead Associate at Novantas, the financial services analytics firm, before landing at Affirm where risk and data science would become his daily vocabulary.
What sets Blumenfeld apart in the crowded fraud-tech landscape is his insistence on specificity. He doesn't write about fraud in general terms. His research dissects the mechanics: how tradelines get built and destroyed, how fraudsters exploited PPP loan programs, how email data can flag identity theft before it happens, why ITINs aren't actually riskier than SSNs in credit card lending despite what most models assume. Each blog post reads less like marketing and more like a practitioner's notebook - someone who has seen the actual data and is documenting what it says.
As COO and Head of R&D, Blumenfeld runs the operational side of SentiLink while simultaneously driving the research agenda. It's an unusual dual mandate. Most companies separate product research from operations - the person worried about quarterly targets rarely has bandwidth to publish independent research. Blumenfeld does both, and the research is part of what gives SentiLink its credibility with the compliance and risk officers who ultimately sign the checks.
SentiLink's product sits at the point of application - the moment a person submits their information to open a new account or take out a loan. Within milliseconds, SentiLink's API cross-references dozens of signals, checking not just whether an identity exists, but whether it behaves the way a legitimate identity should. Synthetic fraud is patient fraud. The fake identities that Blumenfeld and Harris first encountered at Affirm had been built up over years - card after card added as authorized users, small credit lines opened and paid back, credit scores carefully nursed into respectability. Catching them at application time requires knowing what the construction looks like.
Since the Series A in 2019 - a $15 million round led by Andreessen Horowitz - SentiLink has grown steadily into the infrastructure layer of US financial services. The Series B in 2021 raised $70 million from Craft Ventures, with Felicis, a16z, and NYCA joining in. David Sacks of Craft Ventures called it among their fastest-growing portfolio companies. At the time of the round, SentiLink had verified several hundred million applications and served three of the top ten US banks. Revenue had grown 5x in the preceding year. The company had achieved all of that largely through founder-led sales.
Named to Forbes 30 Under 30 in 2020, Blumenfeld has since expanded his public profile through American Banker, where he's argued that federal Customer Identification Program rules - unchanged since they were written to address a different era of fraud - need to be modernized. His argument is straightforward: fraud has evolved; the rules haven't. Regulators can either catch up, or the gap between what fraudsters can do and what the law requires financial institutions to check will keep widening.
He's based in Austin, Texas now - a long way from SentiLink's San Francisco headquarters - which puts him in the middle of the country's second fintech hub while remaining embedded in the company he built. The distance from a Bay Area office hasn't slowed the research output: in early 2026, he published analysis of identity fraud rate trends across the second half of 2025, and in February of that year, SentiLink launched Intercept - precision fraud tools built for the highest-stakes decisions, where the cost of a missed catch is highest and the tolerance for friction is lowest.
The 12 fake people in the Affirm queue started something. They're still being caught, along with millions of others - and Blumenfeld is still running the math.