The IRS Wants Smarter Audits. Palantir Could Help Decide Who Gets Flagged

Summary

The IRS paid Palantir roughly $1.8m last year to improve a custom selection tool called SNAP (Selection and Analytic Platform) that aims to surface “highest-value” audit, collection and criminal-investigation targets from the agency’s maze of legacy systems.

The pilot is intended to sit above more than 100 business systems and hundreds of existing case-selection methods, extracting signals from structured and unstructured data. The IRS asked Palantir to develop three case-selection methods covering disaster-zone claims, Residential Clean Energy Credits (eg, solar installations) and Form 709 gift-tax returns.

Documents show Palantir has been a long-time IRS contractor (over $200m in awards/obligations) but details about how SNAP will integrate with IRS systems, what data it will use beyond “existing data in SNAP today,” and how decisions will be explained to taxpayers remain unclear.

Key Points

The IRS contracted Palantir to upgrade SNAP, a tool to prioritise audits and investigations across fragmented legacy systems.
The pilot targets specific programs: disaster relief claims, Residential Clean Energy Credits, and gift-tax returns (Form 709).
Palantir’s software pulls from unstructured documents and existing data sources, but the IRS restricted the work to data already in SNAP.
The agency has long relied on opaque scoring (eg, DIF scores); SNAP promises modernising case selection but raises transparency, fairness and privacy concerns.
IRS resource constraints, frequent leadership turnover and decades of underfunded technology modernisation complicate large IT projects and oversight.

Why should I read this?

Short answer: because this could change who gets audited. The article shows the IRS is turning to a controversial contractor to help decide which taxpayers are flagged — and it exposes the trade-off between smarter targeting and the risk of black-box decisions. We’ve skimmed the documents and pulled the bits that matter so you don’t have to sift FOIA records yourself. If you care about tax fairness, privacy or how AI-style tools are used by government, this is worth a read.