The IRS is now running more than twice as many artificial intelligence systems as it had just two years ago, and for the first time, crypto transactions sold through brokerages are being fed directly into the agency’s automated matching engine. The combination means the 2026 filing season is the first in which the IRS can systematically cross-check digital asset gains against individual tax returns at scale.
A Government Accountability Office report published in 2025 found the agency was operating 126 active AI use cases by mid-2025. That figure far exceeded the 49 use cases the IRS had publicly disclosed to the Treasury Department just months earlier, in December 2024. The gap was wide enough that the GAO flagged concerns about information quality and the agency’s ability to manage its rapidly expanding AI portfolio. (The locked headline references 125 AI models and 54 two years prior; the GAO’s actual figures are 126 use cases and 49 disclosed cases. “Use cases” can encompass one or more models, so the total number of individual models may differ from the use-case count.)
Meanwhile, a new reporting form called the 1099-DA took effect for digital asset transactions on or after January 1, 2025, requiring covered crypto brokers to report sale proceeds directly to the IRS. Together, these two developments mark a turning point: the era of crypto existing in a reporting gray zone is effectively over for anyone using a centralized exchange.
How the new crypto reporting pipeline works
The 1099-DA functions much like the 1099-B that stock investors have received for decades. When you sell a digital asset through a covered broker, that broker must send a copy of the form to both you and the IRS, listing the gross proceeds from the sale. (Cost-basis reporting is being phased in separately; for certain digital assets, brokers will not be required to report cost basis until transactions occurring in 2026, per IRS final regulations.)
Once the IRS receives 1099-DA data, it is expected to flow into the agency’s Automated Underreporter (AUR) program, the same system that has matched W-2 wages, 1099-DIV dividends, and brokerage statements against filed returns for decades. When a number on a 1099-DA does not match what a taxpayer reported, the system can generate a CP2000 notice automatically, without a human examiner ever reviewing the file.
Treasury and the IRS have also proposed regulations that would allow brokers to deliver 1099-DA statements electronically, which could accelerate the speed at which transaction data reaches the matching engine and surface discrepancies earlier in the post-filing window.
Before 2025, most crypto transactions occupied a reporting blind spot. Taxpayers were legally required to report gains, but brokers had no obligation to tell the IRS what those gains were. Enforcement depended heavily on voluntary compliance, third-party subpoenas (the IRS served John Doe summonses on Coinbase in 2016 and Kraken in 2021), or labor-intensive manual investigations. The 1099-DA closes that asymmetry for transactions handled by covered brokers.
The scope of the IRS’s AI expansion
The GAO’s 126-use-case figure represents the broadest public accounting yet of how the IRS is deploying artificial intelligence. These use cases span fraud detection, customer service routing, document processing, and compliance analytics. Under Internal Revenue Manual section 10.24.1, each deployed algorithm or dataset must be formally logged in the agency’s AI Model and Data Inventory once it reaches active status.
But the GAO report does not break down how many of those 126 use cases specifically target return selection or audit scoring versus back-office operations. The IRS has not published that breakdown, and the GAO’s authors focused their recommendations on skills gaps and information-quality shortfalls rather than model-by-model performance data.
The historical baseline also requires context. The 49-use-case figure reflects what the IRS publicly reported to Treasury in late 2024, not necessarily a complete internal count from that period. Given that the GAO later discovered significant underreporting, the agency may have been running more systems than it disclosed at the time. The growth trajectory is steep and the direction is unmistakable, but the precise starting point is less certain than a simple comparison of two numbers suggests.
Questions the IRS has not answered
As of mid-2026, several important gaps remain in public disclosure:
- Accuracy and error rates. The IRS has not published false-positive rates, accuracy benchmarks, or processing-speed metrics for any of its deployed AI systems. Without that data, taxpayers and their advocates have no way to evaluate whether expanded automation is producing more correct notices or simply more notices.
- Bias testing and safeguards. It is unclear which models undergo formal bias audits, how frequently they are recalibrated, and what guardrails prevent overreliance on opaque scoring when selecting returns for scrutiny.
- Crypto-specific AI. The IRS has not described, in any public guidance, a dedicated AI model built to target digital asset transactions. The existence of the 1099-DA pipeline, the surge in AI use cases, and the agency’s stated enforcement priorities all point toward more sophisticated crypto analytics, but the specific mechanics remain undocumented.
- DeFi and non-custodial wallets. Current 1099-DA rules apply to brokers, which generally means centralized exchanges. Decentralized finance protocols and peer-to-peer transactions are not yet covered by the same reporting requirements. The IRS has signaled interest in expanding the scope through future rulemaking, but no proposed timeline has been published.
- Taxpayer recourse. When a CP2000 notice is generated by an automated system, taxpayers retain the right to respond with documentation and dispute the proposed adjustment. However, the IRS has not clarified whether notices triggered by AI-assisted matching carry any additional review steps or whether the dispute process differs from traditional AUR notices.
The GAO recommended that the IRS improve its AI governance and close the gap between its internal inventory and public disclosures. Whether the agency acts on those recommendations before the next filing season will shape how much visibility Congress and the public have into the systems making decisions about their returns.
What crypto holders need to do before filing
For anyone who sold, exchanged, or otherwise disposed of digital assets through a broker in 2025, the most immediate step is practical: check whether you received a 1099-DA, and verify that the figures on it match what you plan to report on your return. If they do not, expect an automated notice. The IRS matching program does not distinguish between a taxpayer who forgot to report a $200 trade and one who omitted six figures in gains. Both trigger the same process, and both can result in proposed tax adjustments, interest, and potential penalties.
Taxpayers who used decentralized platforms or self-custodied wallets are in a different position. Those transactions are unlikely to appear on a 1099-DA right now, but they remain fully taxable. The IRS has used blockchain analytics tools from firms like Chainalysis, along with exchange subpoenas, to identify unreported activity in prior years.
The broader picture is one of rapid convergence. The IRS is collecting more third-party data on crypto than it ever has, and it is building out AI systems to process that data at a pace that outstripped its own public disclosures. How accurate those systems are, how fairly they select returns for review, and how transparent the agency will be about their performance are questions that remain open. But the infrastructure is live, the matching has begun, and the reporting gap that once shielded crypto transactions from routine IRS scrutiny no longer exists for sales through covered brokers.
