AI system
- System
-
IRS AI and machine-learning case-selection models (two partnership prioritization models; a refundable-credit audit selection model in pilot; machine learning for partnership, income and international tax risk)
- Model
-
The Large Partnership Compliance model for tax year 2021.
- Release
-
Exact release not established.
- Application developer
-
Internal Revenue Service (in-house data scientists with tax-enforcement experts) - Role
-
The models prioritized tax returns for human examiners.
- Configuration
-
The IRS used its Large Partnership Compliance model for tax year 2021.
- Inputs and outputs
-
Not assessed
- Autonomy
-
Not assessed
- Workflow
-
Inflation Reduction Act funding for IRS compliance work. Collaboration between IRS data scientists and tax-enforcement experts. Partnership return data for the Large Partnership Compliance program launched in 2021.
- Human oversight
-
Not assessed
- Relation to AI
-
The models prioritize or select returns for human examiners. Examinations and any assessments are human decisions. The IRS release is the deployer's own account of a system it builds in-house. The GAO post is an independent government source that relays IRS officials' claims about the pilot model and adds caveats. Neither reports an outcome attributable to the AI.
- AI contribution
-
The models prioritize or select returns for human examiners. Examinations and any assessments are human decisions. The IRS release is the deployer's own account of a system it builds in-house. The GAO post is an independent government source that relays IRS officials' claims about the pilot model and adds caveats. Neither reports an outcome attributable to the AI.
Benefits
- Outcome identifiers
-
Not assessed
- Benefit
-
IRS (8 September 2023): AI to help detect cheating and select cases; exams planned for 75 partnerships averaging over $10bn in assets. GAO (Jun 2024): two AI models prioritize partnership audits.
- Category
-
Improved Organizational Efficiency
- Status
-
Not assessed
- Beneficiary groups
-
People. Organizations. Society.
- Beneficiaries
-
Not assessed
- Reach
-
Not assessed
- Distribution and access
-
Not assessed
- Duration
-
Not assessed
- Expected?
-
Yes
- Costs and harms
-
Not assessed
Measurement
- Studies
-
Not assessed
- Study design
-
Not assessed
- Measurement
-
Not assessed
- Setting and sample
-
Not assessed
- Measures
-
Not assessed
- Comparison
-
Not assessed
- Results
-
Not assessed
- Additional value from AI
-
Not assessed
- Limitations
-
Not assessed
- Denominators
-
Not assessed
- Materials for replication
-
Not assessed
Money
- Kind of value
-
additional tax assessments yearly that IRS management estimates from reassigning 14 classifiers to examinations after the Form 1040 return classification AI model automated classification (an SB/SE individual-return model, not the partnership models)
- Amount
-
Estimate: approximately $26.1 million
- Currency and prices
-
Not assessed
- Period and unit
-
Not assessed
- Realization
-
Not assessed
- Recipients and payers
-
the Internal Revenue Service (U.S. Treasury)
- Gross or net
-
Not assessed
- Included costs
-
Not assessed
- Comparison
-
Not assessed
- Method
-
Not assessed
- Return and payback
-
Not assessed
People and places
- Participants
-
Developer:
Internal Revenue Service (in-house data scientists with tax-enforcement experts). Operator:
Internal Revenue Service. - Jurisdictions
-
Benefit:
United States (national). Developer:
United States. Operator:
United States. - Ownership and control
-
Not assessed
- Dependencies
-
Not assessed
- Cross-border connections
-
Not assessed
- Geopolitical implications
-
Not assessed
Follow-up
- Conditions for replication
-
Sustained funding and staffing for the examinations the models select. Documentation of the models, which GAO found lacking in 2024. Correction of the design weaknesses GAO said may limit objective identification of high-risk returns.
- Replication attempts
-
Not assessed
- Later findings
-
19 May 2025. TIGTA's May 2025 audit found the Large Partnership Compliance model still in production, with the first 82 high-risk TY 2021 returns under examination, no closed examination results yet, and uncertainty over reduced staffing after the end of the contractor's work.
- Availability
-
Not assessed
- Institutional decision
-
Not assessed
- Grounds
-
Not assessed
- Action and result
-
Not assessed
- Success criterion
-
Not assessed
Evidence
- Source numbers
-
AIBD-0338-source-0001–0005
- Citations
-
Numbered references below.
- Language
-
Not assessed
- Source relationships
-
Developer, Independent
- Original evidence
-
Not assessed
- Access and versions
-
Not assessed
- Supported findings
-
Not assessed
- Evidence links
-
Not assessed
- Evidence strength
-
Not assessed
- Missing evidence
-
Not assessed
Grades
- Assessed benefits
-
People. Organizations. Society.
- Depth
-
Not assessed
- Reasons
-
People: No individual outcome is documented. Organizations: The IRS gives no results from the AI-selected examinations and GAO reports only officials' pilot claims. Society: No revenue or fairness outcome is attributed to the models.
- Observed or intended
-
Not assessed
- Rubric
-
Not assessed
- Assessor and date
-
Not assessed
- Alternative assessments
-
Not assessed
- Fill
-
Independently studied
Entry details
- Kind
-
Not assessed
- Related entries
-
Not assessed
- Field
-
Public administration: tax enforcement. National revenue agency.
- Purpose
-
Built to select high-risk returns for examination and to detect compliance threats. The claimed benefit is that purpose, with no attributable outcome documented.
- Dates
-
Date basis unassessed: 2023 (precision unassessed). Updated: 21 October 2026.
- Stage
-
Not assessed
- Version
-
SHA-256: 60071e95f8b4…
References
- Other identifiers
-
Not assessed
- Category mappings
-
Not assessed
- Brief
-
Not assessed
- Revisions
-
5 revisions. Latest: 21 October 2026.
Sources
-
Report Independent author
U.S. Government Accountability Office (2024). Artificial Intelligence May Help IRS Close the Tax Gap. GAO WatchBlog, 6 Jun 2024.
IRS currently uses two AI models to help prioritize partnership returns for audit
may limit these models' ability to objectively identify the most high-risk returns
-
Report Independent author
Treasury Inspector General for Tax Administration (2025). The IRS Could Leverage Examination Results in Artificial Intelligence Examination Case Selection Models and Improve Processes to Evaluate Performance. Report 2025-308-022, 19 May 2025.
-
Press release Developer
Artificial Intelligence that will help IRS compliance teams better detect tax cheating, identify emerging compliance threats and improve case selection tools
cutting-edge machine learning technology to identify potential compliance risk in the areas of partnership tax, general income tax and accounting, and international tax