In the 2025 federal AI use case inventory compiled by the U.S. Office of Management and Budget (OMB), agencies reported 3,611 individually described AI use cases. Counting only the 2,959 that are in pre-deployment, pilot, or deployed status, classical machine learning is the most common technique (1,112), ahead of generative AI (853), natural language processing (452), and computer vision (291). By subject, science leads with 755 use cases, followed by administrative functions, information technology, law enforcement, and health and medical work.
Key takeaways
- The inventory is self-reported by 41 agencies that filed individual use cases, and it excludes national security systems and the intelligence community.
- Among 2,959 active use cases, classical machine learning accounts for 37.6 percent and generative AI for 28.8 percent, according to a count of the published data file.
- Science is the largest subject at 755 use cases, and most of those come from NASA, the Department of the Interior, and the Department of Energy.
- Generative AI shows up mainly in information technology and administrative work, while computer vision clusters in health, science, and law enforcement.
- The 2025 inventory is the latest available as of October 2026; counts show how many use cases agencies listed, and some fields are blank or inconsistent.
What the inventory covers and leaves out
Federal agencies must inventory their AI use cases each year under Executive Order 13960, the Advancing American AI Act, and OMB Memorandum M-25-21. OMB publishes the consolidated result in a public repository. Its summary, dated April 13, 2026, counts 56 agency submissions: 41 with individually reported use cases, 46 with consolidated reporting on commercial off-the-shelf AI products, and two agencies, the Consumer Financial Protection Bureau and the Federal Maritime Commission Administration, that reported no AI use. Agency submissions were due to OMB on December 22, 2025, and public reporting was due by January 28, 2026.
The exclusions matter for interpretation. AI used as a component of a national security system or within the intelligence community is outside the inventory, as are Department of War use cases, basic and applied research with limited exceptions, and any details that law or governmentwide policy bars from release. A count of 3,611 therefore describes civilian federal reporting and leaves out part of federal AI.
As of October 27, 2026, this 2025 inventory is the most recent one. OMB's list of repositories shows the 2025 inventory (last updated May 14, 2026) and the earlier 2024 inventory, and no 2026 edition. OMB's guidance describes the inventory as an annual exercise, so another cycle is expected, but no dates for it were found. Figures from the 2025 file should not be read as a 2026 snapshot.
The six techniques agencies can choose from
Each use case carries one classification, which agencies choose as the option that most closely aligns with the system, even though a use case may rely on several. The categories come from OMB's data dictionary and give a practical short list of what the main branches of AI look like in government work. The table counts the 2,959 use cases whose development stage is pre-deployment, pilot, or deployed, using the published CSV file downloaded on October 27, 2026. The counts and shares are the writer's tally of that file.
| Classification | OMB description (shortened) | Use cases | Share |
|---|---|---|---|
| Classical ML | Models trained on data to predict or classify from patterns | 1,112 | 37.6% |
| Generative AI | Generates new content such as text, images, audio, or code | 853 | 28.8% |
| Natural language processing | Processes, interprets, and shares information in human language | 452 | 15.3% |
| Computer vision | Processes and interprets images and video | 291 | 9.8% |
| Agentic AI | Performs tasks or decisions autonomously with minimal human intervention | 117 | 4.0% |
| Other | Not covered by the options above | 102 | 3.4% |
| Reinforcement learning | Trained by trial and error with rewards and penalties | 12 | 0.4% |
| Blank | No classification entered | 20 | 0.7% |
Generative AI is the most uneven row. Of its 853 use cases, 447 are still in pre-deployment, 154 are pilots, and 252 are deployed. Only about three in ten generative AI entries (252 of 853) are deployed, so most are plans or trials.
What the subjects look like
Each use case also carries a topic area, again chosen by the agency. Among the same 2,959 use cases, the largest are science (755), administrative functions (420), information technology (381), law enforcement (285), and health and medical (272). Service delivery follows with 147, procurement and finance management and energy and environment with 90 each, and transportation with 79. The "Other" category appears under several spellings, so the smaller categories are split into variants that a reader can merge.
The science total depends on a few agencies. NASA reports 384 active science use cases, the Department of the Interior 170, the Department of Energy 84, and the Department of Agriculture 60. Of the 755 science use cases, 532 are classified as classical machine learning, so most science entries are prediction or classification models rather than chat tools.
Where techniques meet subjects
- Generative AI: information technology (222) and administrative functions (206) lead, ahead of science (63) and service delivery (62).
- Computer vision: health and medical (76), science (74), and law enforcement (69).
- Natural language processing: administrative functions (84), health and medical (71), and law enforcement (54).
- Classical machine learning: science (532), law enforcement (101), and administrative functions (78).
The split follows the data each tool handles. Generative tools cluster where text, code, and paperwork are the product, and vision models appear where images are the data, from medical scans to field imagery.
High-impact use cases
OMB's inventory flags 445 use cases as high-impact. The repository's agency table shows the Department of Veterans Affairs with 215, the Department of Justice with 114, and the Department of Homeland Security with 55. Counting the file by subject, health and medical accounts for 140 of the flagged use cases and law enforcement for 130, with 55 carrying no topic. For flagged cases the data file carries extra columns covering testing, assessment, independent review, monitoring, training, fail-safes, and appeals.
Reading the counts with care
- A use case is an entry on a list, not a measure of volume. One entry may describe a pilot with a few users and another a tool used by thousands of employees.
- Agencies decide how to divide their work into entries, so a large count can reflect detailed reporting as easily as heavy use.
- Some fields are incomplete. In the file, 338 records have a blank development stage and 603 have a blank topic area, mostly retired entries or entries with a blank stage.
- The repository summary lists 1,818 deployed and piloted use cases, while the stage column shows 1,480 as deployed or pilot. The gap equals the 338 blank-stage records, and the summary table treats all of them as deployed or piloted: 223 belong to Commerce, 59 to the Tennessee Valley Authority, and 56 to Education.
- The file is updated on a rolling basis, so later downloads can differ from the figures here.
FAQ
What are the main branches of AI in this data?
OMB's classification offers agentic AI, classical or predictive machine learning, computer vision, generative AI, natural language processing, and reinforcement learning. Classical machine learning has the most active use cases, and generative AI is second.
What are the main application areas?
By agency-chosen topic, science, administrative functions, information technology, law enforcement, and health and medical are the five largest among active use cases.
Does the inventory show how much the government uses AI overall?
No. It lists civilian use cases that agencies reported, leaves out national security and intelligence use, and does not measure usage volume.
Why do the totals differ between the repository summary and the file?
The summary counts 3,611 use cases across all stages, which matches the number of rows in the file. The deployed and piloted total differs because some records have a blank stage field.
Where can the data be checked?
The OMB repository publishes the CSV and Excel files, a data dictionary, and the reporting guidance, and each agency also hosts its own inventory.
Related reading
- Key topics in artificial intelligence
- What neural networks are used for
- Deep learning applications
- Why AI is considered beneficial
More in AI Basics.
Last verified: October 2026
Sources
- 2025 Federal Agency AI Use Case Inventory (README, data files, data dictionary), Office of Management and Budget, repository summary dated April 13, 2026
- Federal Agency AI Use Case Inventory repositories list, Office of Management and Budget, checked October 2026
- Guidance on 2025 Agency Artificial Intelligence Reporting, Office of Management and Budget, 2025
- Memorandum M-25-21, Accelerating Federal Use of AI through Innovation, Governance, and Public Trust, Office of Management and Budget, 2025
- Executive Order 13960, Promoting the Use of Trustworthy Artificial Intelligence in the Federal Government, December 2020