Federal AI Intelligence
Menu
Snapshot · OMB 2025 inventory · processed Oct 11, 2026
Technology landscape

Technology landscape

How reported AI classifications spread across agencies, missions, lifecycle stages and sourcing. Each record carries one reported classification, so categories do not overlap. A classification describes the technique, not a specific commercial model.

Agency
Stage
High-impact
Topic
AI type
Sourcing
Vendor
Function
Start year

Only records with a parseable start date match a year filter.

Lifecycle by technology

Share of each technology's records
DeployedPilotPre-deploymentRetiredNot reported
Generative AI858
Classical / predictive ML1,114
Natural language processing454
Computer vision292
Agentic AI117
Reinforcement learning12
Other102
Not reported662

Sourcing by technology

Where reported
Developed in-houseContract and in-housePurchased from vendorNot reported
Generative AI858
Classical / predictive ML1,114
Natural language processing454
Computer vision292
Agentic AI117
Reinforcement learning12
Other102
Not reported662

Technology adoption by agency

Row share: each agency's records by classification · 16 largest portfolios in the filter
Agency by AI classification
Generative AIClassical / predictive MLNatural language processingComputer visionAgentic AIReinforcement learningOtherNot reportedn
HHS31%19%26%4%6%2%11%447
NASA9%66%6%11%5%1%1%0%425
VA15%21%16%22%7%0%20%367
DOE44%31%6%2%4%1%7%4%340
DOJ28%34%20%10%1%6%314
DOI6%69%6%8%2%0%3%5%247
DHS26%24%13%21%1%14%238
DOC100%223
USDA20%48%9%10%2%12%162
TREAS60%13%12%2%2%2%9%129
DOT21%7%4%6%61%70
SEC40%23%3%7%5%2%20%60
STATE30%10%17%2%3%38%60
TVA100%59
ED100%56
FDIC14%30%18%4%6%28%50

Technology by topic area

Row share within each topic
Topic by AI classification
Generative AIClassical / predictive MLNatural language processingComputer visionAgentic AIReinforcement learningOtherNot reportedn
Science8%70%6%10%4%1%1%758
Administrative Functions48%19%20%6%3%0%3%1%430
Information Technology57%18%13%2%7%2%3%394
Law Enforcement19%35%19%24%2%1%285
Health and Medical17%25%26%28%4%1%272
Other33%28%25%3%1%0%9%0%223
Service Delivery43%23%26%5%2%1%1%150
Energy and the Environment13%37%4%7%2%2%4%31%131
Procurement and Financial Management40%33%16%1%9%1%90
Transportation13%22%9%9%4%1%43%79
Government Benefits Processing34%27%16%7%16%56
Cybersecurity26%53%9%2%8%2%53
Human Resources57%17%15%7%2%2%46
Emergency Management12%48%24%9%3%3%33

High-impact designation by technology

High-impactPresumed high-impact, determined notNot high-impactNot reported
Generative AI858
Classical / predictive ML1,114
Natural language processing454
Computer vision292
Agentic AI117
Reinforcement learning12
Other102
Not reported662

Governance disclosure by technology

High-impact records: reported in place / scope
TechnologyHigh-impactTestingImpact assess.Indep. reviewMonitoringTraining
Generative AI603 / 604 / 604 / 602 / 605 / 60
Classical / predictive ML13219 / 13216 / 13216 / 13219 / 13218 / 132
Natural language processing564 / 562 / 562 / 562 / 562 / 56
Computer vision11819 / 11815 / 11815 / 11817 / 11817 / 118
Agentic AI220 / 220 / 220 / 220 / 220 / 22
Reinforcement learning0–––––
Other20 / 20 / 20 / 20 / 20 / 2
Not reported550 / 550 / 550 / 550 / 550 / 55

Shows counts reported in place against all high-impact records of that technology. The remainder includes in-progress, not applicable and unanswered. Full governance view

Systems, vendors and technology are different things

790 of 3,611 records name a vendor (22%)
Microsoft133133 recordsOpenAI4444 recordsGoogle3434 recordsAmazon Web Services2727 recordsDeloitte2727 recordsPalantir2626 recordsServiceNow1919 recordsThomson Reuters1515 recordsLexisNexis1111 recordsNEC1111 recordsLeidos1010 recordsGuidehouse99 recordsAccenture77 recordsAdobe77 recordsAretec77 records

Vendor: the supplier the agency names in a free-text field. Matched to a standard list of companies; one record may name several.

System name: the agency's name for the system (reported for 25% of records).

AI classification: the technique category the agency selected. A vendor name does not show which model or technique is used, and this platform does not infer it.