What the IRS disclosed
Every IRS system that handles taxpayer data has to have a Privacy and Civil Liberties Impact Assessment, and the IRS posts them on IRS.gov. One of them, approved on March 10, 2025, is for a tool called the Backup Withholding Selection Assistance Tool RPA. RPA stands for robotic process automation: software that operates the same screens and spreadsheets a person would. It belongs to the Small Business/Self-Employed division and was funded as Inflation Reduction Act initiative 4.3.
The assessment states the problem in the IRS's own words. Payers who file 1099s with missing or incorrect payee TINs "are required to pay 24% backup withholding tax, but per the Treasury Inspector General for Tax Administration (TIGTA), noncompliance may result in a significant tax gap, of billions of dollars potentially each year." The tool's job is case selection: it "identifies, selects and delivers to the campus, an optimal mix of cases to promote the highest degree of voluntary compliance, and to ensure fair coverage."
Backup Withholding Selection Assistance Tool RPA, Privacy and Civil Liberties Impact Assessment (approved Mar. 10, 2025).
How the robot selects cases
The assessment describes the process in some detail. The automation "will select the relevant backup withholding cases most likely to result in the successful recovery of unpaid taxes, generate files with details of approximately 100,000 TINs, and narrow the list using specified criteria." It "compares data by Taxpayer Identification Number (TIN) from the Compliance Data Warehouse to corresponding Integrated Data Retrieval System (IDRS) data to identify case selection criteria."
Sources: BWH RPA assessment (2025); Payer Master File assessment (Aug. 25, 2026); Information Returns Database assessment (June 8, 2026).
Four details from the assessment are worth knowing:
- It is attended. The robot runs under an IRS analyst's own profile, and the analyst signs in to IDRS as usual, so it sees only what that analyst is entitled to see. A person starts it and receives its output.
- It is not machine learning. Asked whether the tool is "trained through the use of algorithms," the IRS answered no. It automates a fixed extraction and comparison that people used to do by hand in Excel and IDRS.
- It works on businesses. The assessment says EINs "are required to identify the businesses whose cases are available for selection," with names and addresses for case processing. The payer is the target; payee SSNs and ITINs are in the data because they are on the 1099s.
- The output is kept. Result files go to an access-restricted shared drive and fall under the Compliance Data Warehouse records schedule: deleted ten years after the data update, or later if still needed.
Why the IRS built it
Selection used to be the bottleneck. In the assessment's words, the robot "performs a data extraction from, collection and comparison process that takes a team of 5 people about 2 weeks to manually complete," does it "in minutes," "improves accuracy, and most importantly provides valuable human resources for other priorities."
The background is two audits by the Treasury Inspector General. For tax year 2018 TIGTA counted 182,075 payers that filed 440,404 information returns with missing or incorrect TINs and did not withhold $13.3 billion on $55.6 billion of payments. The IRS's response in 2019 was a backup withholding group of 13 tax examiners, and TIGTA's 2021 recommendation was that the IRS use "data analytics ... to maximize its limited resources by focusing enforcement efforts on highly noncompliant payers." The robot is that recommendation put into practice at the selection step. The TIGTA findings page has the full numbers.
The robot does not send letters or assess tax. It decides who gets looked at. The examiners in the campus backup withholding program still work each case under IRM 4.19.26, starting with Letter 6112.
Where your 1099s enter it
The robot reads the Compliance Data Warehouse, which "captures data from multiple production systems" and organizes it for analysis. Two of those production systems hold what you file, and both have their own published assessments:
| System | What it holds about you | What the IRS says it is used for |
|---|---|---|
| Payer Master File (assessment approved Aug. 25, 2026) | A composite of every information return you submit, with your Forms 1096, extensions and waiver requests, kept ten years (five active, five in a retention file) | Comparing what payers report with what payees file, and the civil penalty program: identifying payers "who file information returns late, submit paper returns when electronic filing is required, or report incorrect Taxpayer Identification Numbers." One of its listed outgoing interfaces is to the information returns underreporter and backup withholding system. |
| Information Returns Database (assessment approved June 8, 2026) | Validated 1099 data received from IRIS since 2023, alongside ACA returns | Storage and distribution to downstream systems, including the Compliance Data Warehouse, for "compliance, reporting, filing verification, withholding validation, fraud detection" |
The practical point: the evidence the robot works from is your own filing. A 1099 with an empty TIN box, or with an amount in box 4 and no Form 945 under the same EIN, is a record in these systems from the day it posts. With FIRE closing on November 19, 2026, every 1099 filed from 2027 on arrives through IRIS. The FIRE to IRIS page covers that change.
What it selects on, and what is not public
The assessment does not list the "specified criteria" used to narrow the 100,000 TINs, and nothing else the IRS has published does either. The IRM for the program also keeps its case tolerance in a redacted subsection. This guide does not guess at either.
What is public is the kind of case the program works. IRM 4.19.26 describes two selection workstreams: payers who filed 1099s or W-2Gs with no payee TIN and no Form 945, and payers who reported withholding in box 4 and filed no Form 945. The stated aim of the robot, cases "most likely to result in the successful recovery of unpaid taxes," points toward the payers with the largest unwithheld amounts, which is also where TIGTA told the IRS to look. The Form 945 compliance check guide walks through both workstreams, the letters and what the IRS accepts.
What it means for a payer
- Assume the list gets built every cycle. When selection took two weeks of five people's time, it was rationed. At minutes per run, the constraint is examiner time to work the cases, not the IRS's ability to find them.
- Treat a blank TIN on a filed 1099 as a 24% liability. For a missing TIN, backup withholding is due at the time of payment, with no notice and no dollar threshold. Compute your own exposure: the total on every 1099 you filed with an empty TIN box, times 0.24. The missing-TIN exposure calculator does it from your numbers.
- Reconcile box 4 with Form 945 before you file. Any box 4 amount without a Form 945 under the same EIN is the second workstream. Fix data-entry errors in box 4 before the 1099 is furnished, and file the 945 where the withholding is real.
- Keep the date each TIN arrived. The defense, if you are selected, is proof that the TIN was on file when the payment was made. A W-9 dated before the payment answers the case; a TIN collected after the letter does not.
- Keep your Payer Master File address current. Letters go to the address the IRS has for your EIN. File Form 8822-B when you move.
Questions people ask
Is this AI deciding my case?
No. It is robotic process automation: software that repeats a fixed set of steps in Excel and IDRS that analysts used to do by hand. The IRS says it is not trained on algorithms, it runs only when an analyst starts it, and it selects cases rather than deciding them. An examiner works every selected case.
Will the letter say I was picked by the robot?
No. A selected payer receives the same Letter 6112 from the campus backup withholding program, with a list of the payees in question. The letter does not say how the case was chosen, and the answer to it is the same either way.
Does the robot use IRS TIN Matching results?
Not according to the assessment, which lists Microsoft File Explorer, Excel, IDRS and an IDRS interface as the systems it uses. TIN Matching is a payer's own pre-filing check; nothing published says the IRS looks at whether you used it when selecting cases.
Does it look at incorrect TINs as well as missing ones?
The assessment's summary mentions both. For an incorrect TIN, though, backup withholding starts only after a CP2100 and the 30 business days to act on it, so the clearest liability, and the one the published workstreams target, is the missing TIN and the unreported box 4 withholding.
Where can I read the document?
On IRS.gov: the Backup Withholding Selection Assistance Tool RPA assessment (PDF). Names of the IRS staff involved are withheld as For Official Use Only.