Alex Shieh is using AI to hunt $600B in government fraud — and gets paid only if he wins

Alex Shieh, Anti-Fraud AI

Alex Shieh dropped out of Brown in 2025, did a stint at Palantir, and is now building what he describes as "the Palantir of fraud detection" — a contingency-fee business that uses AI to identify government fraud and then sues the fraudsters under federal whistleblower laws.

The model is straightforward. Anti-Fraud AI keeps 15 to 30% of any recovery it wins for the government; if it loses, it earns nothing. Shieh estimates $600 billion a year — roughly 8% of the federal budget — is lost to fraud across government programs. The company is currently filing cases across defense, small business administration loans, and healthcare, where he notes the government spends in the trillions.

“We use we're building AI models to detect fraud when it happens, and then we go and sue the fraudsters and we only make money on contingency when we actually get a recovery for the government. We're able to keep 15 to 30% under these whistleblower laws. There's $600,000,000,000 in fraud — you're not gonna get tips for most of it.”

No revenue for years

The catch is timing. The U.S. legal system moves slowly, and Shieh expects three to four years before Anti-Fraud AI sees its first dollar of revenue. That's a long runway to fund on conviction alone, and the 11-person team is actively hiring AI engineers.

How the detection works

Shieh describes a two-layer architecture. The first layer uses LLMs to extract structured entities from unstructured documents — contracts, filings, databases — and assembles them into a knowledge graph. The second layer applies a rules model developed with input from his two lawyer co-founders, who bring legal expertise on what fraud actually looks like. The system flags potential violations in real time and routes them to human investigators for follow-up.

The starting point is open-source intelligence — publicly available data that lets the team cast the widest net before layering in closed sources, FOIA requests, or purchased datasets. The analogy Shieh uses: fraud always leaves traces, but they're scattered across different data sources, like blind men touching different parts of an elephant. The job is connecting them.

Timing and tailwinds

Shieh argues the anti-fraud moment is genuinely political right now, pointing to DOGE's public scrutiny of government spending and journalist Nick Schirle's investigation into Minnesota daycare centers allegedly defrauding Medicaid. Both, he says, have driven awareness and inbound interest in Anti-Fraud AI. He also notes that Sacramento is reportedly considering legislation in response to Schirle's reporting — a sign of how politically charged the space has become.

The business only works at scale with AI. Historically, fraud cases came from insider tips or luck. Shieh's bet is that systematic, outbound detection across the full universe of government spending is only possible if you automate the discovery layer — and that $600 billion is too large a target to rely on tips alone.