Palantir tests a system for the IRS to select priority targets for tax audits

The US Internal Revenue Service is testing a Palantir tool that helps select the most beneficial audits and investigations for the government. The pilot should collect signals from dozens of old IRS systems and more quickly find cases where tax debt, fraud, or grounds for criminal prosecution are possible.

How the pilot works

Based on documents obtained through freedom of information requests, the IRS paid Palantir $1.8 million in 2025 for developing the Selection and Analytic Platform, or SNAP. This tool should improve the selection of priority cases for audit, collection of overdue taxes, and potential criminal investigations.

For the service itself, the task looks pragmatic: instead of manually wandering through an old technology stack, gather the picture in one place and spend less time on audits that give the budget almost nothing. The problem is that the IRS has accumulated very heavy infrastructure. In the documents, the service acknowledges that it uses more than 100 business systems and about 700 selection methods that have been built up over decades.

SNAP doesn't replace this entire landscape in one step, but covers it from above: it helps auditors see signals from fragmented databases and search for important details in unstructured attachments that were previously harder to correlate with each other. According to government procurement data, Palantir has been working with the IRS for more than ten years, and the total volume of contracts and payment obligations has exceeded $200 million.

Which cases are in focus

In the first phase, the IRS asked Palantir to develop three separate case selection methods tied to specific sections of the tax code. This is not an abstract search for "suspicious citizens," but a quite applied pilot with clear categories, where the service wants to test whether the new approach helps find more productive cases. This design is important: the agency is limiting the experiment and can compare the new logic with old procedures on specific types of statements and declarations.

applications for tax benefits for residents of disaster zones Residential Clean Energy Credit - benefits for installing solar panels, wind turbines, and other home energy equipment Form 709 - declarations for gift tax, when shares, businesses, works of art, and other valuable assets are transferred supporting documents that can clarify asset valuation, relationships between parties, and calculation logic Unstructured data is particularly interesting here. For gift tax cases, these could be appraisal documents, balance sheets, information on business income, and descriptions of relationships between the donor and recipient. However, in the contract materials it is separately stated that Palantir should work with data already existing within SNAP, rather than pulling new external data streams into the system at its discretion.

It is precisely in such files that signals often hide that are not in the standard declaration fields.

Why this is controversial The IRS already has algorithmic selection mechanisms.

For decades, the service has relied on the DIF score - an internal metric that assesses the probability that a declaration deserves examination. How exactly it is calculated is not publicly disclosed, and researchers have long called this approach a black box. SNAP takes the next step: not just ranking declarations, but helping to link more sources together and raise cases that an inspector might have missed.

The risk here is not only in privacy but also in the quality of solutions. In May 2025, a TIGTA audit stated that the IRS is already using AI models to select audits and wants to reduce the burden on honest taxpayers, but has not yet fully established an evaluation of the effectiveness of such models compared with old methods. That is, the service wants to more accurately hit violations, but it still needs to prove that the new models really work better and don't create additional errors.

"The IRS has basically never accomplished a full modernization since the 1960s," says Professor Erika Neumann.

This phrase well explains why the service is turning to an external contractor at all. IRS modernization has stalled for many years, and in 2025 the agency also lost tens of thousands of employees due to layoffs, delayed departures, and early retirements. When you have fewer people and too many fragmented systems, the temptation to hand over case selection to a smarter analytical layer becomes almost inevitable. And this explains the interest in systems that promise to compress weeks of manual analytics into hours.

What this means

For Palantir, this is another chance to embed itself more deeply in critical US government infrastructure. For the IRS, it is an attempt to turn fragmented tax data into a more precise audit selection mechanism. If the pilot shows results, similar systems will begin to influence not only the pace of audits but also who the government considers a priority target for oversight. The next debate will no longer be about whether such software is needed, but about the rules for testing it and accountability.