Report
AI across financial services
AI language appeared more often in annual financial filings, while private-offering notices and a Census survey provide separate evidence on issuer names and business use.
Euphemios12 min read
Executive summary
AI wording appeared more often in original annual reports across all five financial sub-industries tracked here. The largest increase was among banks. 80.79% of bank filings through November 20, 2026, matched the phrase artificial intelligence, compared with 59.65% through the same day in 2025. In a separate survey, 42.7% of in-scope finance and insurance businesses reported using AI or what the Census Bureau now calls Super Intelligence in any business function, compared with 25.0% across all surveyed sectors. The latest survey asked about the two weeks ended November 1.
Private-offering notices tell a different story. 8,050 hedge-fund notices were in the latest three-quarter Form D data, and 11 of 7,637 distinct issuer identifiers matched the AI-word name screen. Separately, 7 issuers in the specified financial-services groups matched the name screen during those quarters. The three measures cover different populations and answer different questions.
Key findings
- The phrase appeared in 207 of 347 bank filings in 2025 and in 265 of 328 in 2026 over the equal-cutoff periods. The share rose 21.14 percentage points as the filing count fell. (SEC EDGAR full-text search, November 20, 2026.)
- Investment advice had the highest current phrase-match rate among the five SIC groups, at 91.67%, or 55 of 60 filings. Insurance carriers and agents reached 89.93%, and brokers, dealers and exchanges reached 83.87%. (SEC EDGAR full-text search.)
- The stated alternative-manager list matched in 7 of 7 filings, compared with 9 of 10 in the fintech and payments list and 6 of 7 in the crypto list, through November 20, 2026. (SEC EDGAR full-text search.)
- 8,050 hedge-fund Form D notices appeared in the first three 2026 quarters, compared with 7,937 in the same 2025 quarters. Their initial notices with a numeric amount summed to $6.87 billion, with blank and indefinite amounts excluded. (SEC Form D Data Sets.)
- 7 AI-named issuers in financial-services groups had $106.27 million in reported numeric offering amounts on their initial notices through the third quarter. One notice accounted for $100 million and reported no amount sold when filed. (SEC Form D Data Sets and the original notice.)
- Finance and insurance reported 42.7% current AI or Super Intelligence use, 17.7 percentage points above the survey's national estimate for the same reference period. (U.S. Census Bureau, Business Trends and Outlook Survey, November 19 release.)
AI wording rose across the five SIC groups
Banks went from 6 of 435 annual filings matching the phrase in 2019 to 265 of 328 through November 20, 2026. In the equal-cutoff periods, the share increased more for banks than for the other four groups. Investment advice led on share in the latest period. Insurance carriers and agents supplied more filings than investment advice, and 125 of 139 matched. Brokers, dealers and exchanges contributed 26 of 31 matches. Other credit and finance was the smallest share of the five groups in 2026 at 71.83%, up from 62.83% through the same date in 2025. The categories reflect the industry codes returned on the filings. Investment advice covers filers with that designation and provides only a partial view of investment management.
AI phrase matches by financial sub-industry
- 2025 through Nov 20
- 2026 through Nov 20
| Returned SIC group | 2025 through Nov 20 | 2026 through Nov 20 |
|---|---|---|
| Banks | 59.654179 | 80.792683 |
| Brokers, dealers and exchanges | 79.310345 | 83.870968 |
| Investment advice | 81.632653 | 91.666667 |
| Insurance carriers and agents | 82.857143 | 89.928058 |
| Other credit and finance | 62.831858 | 71.830986 |
Share of original 10-Ks matching the phrase artificial intelligence. Both windows end November 20. A mention can appear in a risk, project or other disclosure.
Source · U.S. Securities and Exchange Commission EDGAR full-text search; Euphemios analysis · Original 10-K filings by filing date, January 2019 through November 20, 2026 · As of
An annual report might discuss an internal project, a competitor, new oversight or a risk that has yet to materialize. Each original filing contributes one observation even if it mentions the phrase repeatedly. A filing rate gives each eligible report equal weight. It can change when the mix of filers changes, and it gives the same weight to a small issuer's annual report as a large issuer's.
What we cannot measure yet. The filing search leaves the number of AI deployments, budgets, productivity gains and use in specific business functions unmeasured. It also misses firms that never file this form and concepts described without the phrase.
Fintech, alternative managers and crypto require stated issuer lists
Fintech and payments have no dedicated SIC group. The stated list covers card networks, payment processors, small-business and consumer financing, digital banking and an app-based brokerage. It recorded 9 of 10 matches through November 20, 2026, compared with 10 of 10 through the same date in 2025. The stated alternative-manager list covers publicly filing managers with strategies such as private equity or credit. All 7 eligible annual reports in that list matched in each comparison period. The financial results of those managers and the holdings of the funds they oversee are separate matters.
AI phrase matches in the stated issuer lists
- 2025 through Nov 20
- 2026 through Nov 20
| Stated issuer cohort | 2025 through Nov 20 | 2026 through Nov 20 |
|---|---|---|
| Fintech and payments | 100 | 90 |
| Alternative managers | 100 | 100 |
| Crypto | 100 | 85.714286 |
Share of eligible original 10-Ks in each stated list. Both windows end November 20. Filings and issuers can enter or leave the period denominator.
Source · U.S. Securities and Exchange Commission EDGAR full-text search; Euphemios analysis · Original 10-K filings by filing date, January 2019 through November 20, 2026 · As of
The stated crypto list includes an exchange, a stablecoin issuer, miners or infrastructure businesses, and a bitcoin treasury company. Its share went from 6 of 6 to 6 of 7 matches. The denominator changed as individual filing calendars and eligibility changed. In the later period Circle Internet Group entered the comparison and CleanSpark fell outside the cutoff. Hut 8 supplied the only nonmatching original in the later period. These cohorts can overlap the industry groups and should be read on their own.
What we cannot measure yet. These lists have no statistical weighting and leave most private firms out. The reports of listed alternative managers provide no systematic count of AI use among the hedge funds they manage. The crypto filings provide no measure of activity by autonomous software on exchanges or networks.
Hedge-fund notices show offerings and very few AI-word names
Form D is a notice of an exempt securities offering. The hedge-fund series counts original and amended notices where the issuer selected pooled investment fund and hedge fund. 8,050 notices through the third quarter of 2026 came from 7,637 distinct issuer identifiers. 11 matched the name test, or 0.144%. The same quarters in 2025 had 10 name matches among 7,513 issuers, or 0.133%. Multiple notices from the same issuer can enter the filing count. The name share counts an issuer once per period.
Hedge-fund Form D notices by filing period
| Filing period | Hedge-fund notices |
|---|---|
| 2019 | 10017 |
| 2020 | 9738 |
| 2021 | 10312 |
| 2022 | 10260 |
| 2023 | 9778 |
| 2024 | 9871 |
| 2025 | 10118 |
| 2025 Q1–Q3 | 7937 |
| 2026 Q1–Q3 | 8050 |
Includes original and amended notices. The last two bars compare the first three quarters in each year.
Source · U.S. Securities and Exchange Commission Form D Data Sets; Euphemios analysis · Quarterly Form D packages from 2019 through the third quarter of 2026 · As of
The $6.87 billion sum of numeric offering amounts through the third quarter of 2026 is based on 69 original notices with a finite amount. Another 1,380 originals gave an indefinite or blank amount. The comparable sum in 2025 was $9.42 billion. The full-year series can swing with a few large filings. For example, the largest numeric amount in 2023 was $24.81 billion of $33.83 billion in posted offering amounts across all qualified initial notices that year. Amendments contribute to the notice count but add no dollars to this sum.
Hedge-fund offering amounts reported on initial notices
| Filing period | Numeric offering amounts |
|---|---|
| 2019 | 17.240037698 |
| 2020 | 26.347547154 |
| 2021 | 8.109137775 |
| 2022 | 159.107771119 |
| 2023 | 33.828958905 |
| 2024 | 16.127905357 |
| 2025 | 13.291255073 |
| 2025 Q1–Q3 | 9.415731953 |
| 2026 Q1–Q3 | 6.86583552 |
Sum of numeric offering amounts. Excludes amendments, blank amounts and indefinite amounts. The last two periods cover the first three quarters.
Source · U.S. Securities and Exchange Commission Form D Data Sets; Euphemios analysis · Quarterly Form D packages from 2019 through the third quarter of 2026 · As of
What we cannot measure yet. The dollar sum leaves indefinite offerings unpriced. The notices give no census of funds, assets under management, strategies using AI, actual cumulative capital raised or investment results. AI in an issuer's name can reflect a strategy, a brand or other naming choice.
AI words in offering names give a narrow view of private finance
The name screen matched 7 distinct issuers classified in the specified financial-services groups through the third quarter of 2026, compared with 2 through the third quarter of 2025. Their original notices with numeric amounts summed to $106.27 million, compared with $1.80 million in the comparable 2025 window. The 2026 amount is heavily concentrated. A single $100 million Investing notice made up about 94.1% of the period total. Without that notice, the other disclosed numeric amounts sum to $6.27 million.
Distinct AI-named issuers in Form D financial-services groups
| Filing period | AI-name matches |
|---|---|
| 2019 | 1 |
| 2020 | 0 |
| 2021 | 0 |
| 2022 | 3 |
| 2023 | 1 |
| 2024 | 7 |
| 2025 | 4 |
| 2025 Q1–Q3 | 2 |
| 2026 Q1–Q3 | 7 |
Case-insensitive legal-name screen. No matching name produced a measured zero. The last two periods cover the first three quarters.
Source · U.S. Securities and Exchange Commission Form D Data Sets; Euphemios analysis · Quarterly Form D packages from 2019 through the third quarter of 2026 · As of
The Form D classifications describe the issuer's chosen industry group. Investing and Investment Banking can include vehicles whose legal names mention AI. Insurance and other financial services can include operating companies. Some financial technology businesses may appear under Technology or use an issuer name unrelated to their trade name. The name-based series retains both kinds of entities that pass the stated group and name rules.
What we cannot measure yet. The screen leaves the size of fintech fundraising and its AI share unknown. An issuer's name tells us little about its product. The offering amount is what the notice reports for the offering and leaves eventual proceeds unresolved.
Technology spending is visible only where tags are comparable
A stated list of 19 large banks, insurers and payments firms yielded 8 issuers with eligible annual tags in both fiscal years examined. Coverage was 5 of 6 for banks, 0 of 6 for insurers and 3 of 7 for payments firms. Fiscal 2025 ended in September for Visa and December for all other firms shown. The comparisons use the company's own full fiscal years.
Tagged communications and technology expense
- Fiscal 2024
- Fiscal 2025
| Issuer | Fiscal 2024 | Fiscal 2025 |
|---|---|---|
| JPMorgan Chase | 9.831 | 11.029 |
| Citigroup | 9.035 | 9.423 |
| Bank of America | 7.231 | 7.453 |
| U.S. Bancorp | 2.074 | 2.211 |
| Goldman Sachs | 1.991 | 2.17 |
| Visa | 0.778 | 0.894 |
Annual reported operating expense for issuers with the same eligible tag in both years. Fiscal year-ends can differ. Software payments use separate tags.
Source · U.S. Securities and Exchange Commission EDGAR XBRL Company Facts; Euphemios analysis · Annual consolidated dollar facts for fiscal 2024 and 2025, available through November 20, 2026 · As of
Mastercard reported $726 million in fiscal 2025 cash payments tagged for software acquisition, compared with $720 million in fiscal 2024. FIS reported $835 million in fiscal 2025 and $720 million in fiscal 2024 under a broader software-payment tag. The separate software measures and the operating-expense chart describe different accounting lines. The chart's communications and technology line can include obligations for communications, data processing, software or other expenses, depending on the filing.
What we cannot measure yet. The tags leave AI-specific spending, insurer spending under other accounts, spending embedded in third-party service contracts and differences in accounting policy unresolved. The absence of a comparable tag in Company Facts leaves that company's value missing.
Finance and insurance businesses report current use above the national estimate
For the latest period, the U.S. Census Bureau estimated 42.7% in Finance and Insurance and 25.0% nationally. The response covers any business function during the two weeks ended November 1. The national estimate weights businesses across the survey's in-scope sectors. The Census Bureau excludes, among others, central banks and funds, trusts and other financial vehicles. Those exclusions matter particularly when comparing this series with the private-fund and public-filing measures.
Business-reported AI or Super Intelligence use
- Finance and Insurance · AI wording
- National · AI wording
- Finance and Insurance · SI wording
- National · SI wording
| Collection start | Finance and Insurance · AI wording | National · AI wording | Finance and Insurance · SI wording | National · SI wording |
|---|---|---|---|---|
| 2025-11-17 | 29.9 | 17.3 | Missing | Missing |
| 2025-12-01 | 29.4 | 17.2 | Missing | Missing |
| 2025-12-15 | 30.5 | 17.8 | Missing | Missing |
| 2025-12-29 | 30.7 | 17.7 | Missing | Missing |
| 2026-01-12 | 30.3 | 18.2 | Missing | Missing |
| 2026-01-26 | 31.7 | 18.9 | Missing | Missing |
| 2026-02-09 | 30.4 | 17.5 | Missing | Missing |
| 2026-02-23 | 29.6 | 18.2 | Missing | Missing |
| 2026-03-09 | 31.1 | 18.9 | Missing | Missing |
| 2026-03-23 | 34.1 | 19.1 | Missing | Missing |
| 2026-04-06 | 33 | 19.8 | Missing | Missing |
| 2026-04-20 | 33.9 | 19.8 | Missing | Missing |
| 2026-05-04 | 31.3 | 19.5 | Missing | Missing |
| 2026-05-18 | 35.7 | 20.6 | Missing | Missing |
| 2026-06-01 | 34.9 | 20.6 | Missing | Missing |
| 2026-06-15 | 34.3 | 20.6 | Missing | Missing |
| 2026-06-29 | 36 | 21.7 | Missing | Missing |
| 2026-07-13 | 35.5 | 21.5 | Missing | Missing |
| 2026-07-27 | 36.8 | 21.8 | Missing | Missing |
| 2026-08-10 | 40 | 22.4 | Missing | Missing |
| 2026-08-24 | 36.5 | 23.2 | Missing | Missing |
| 2026-09-07 | 41 | 23.8 | Missing | Missing |
| 2026-09-21 | 40.1 | 23.8 | Missing | Missing |
| 2026-10-05 | 40.5 | 24.6 | Missing | Missing |
| 2026-10-19 | Missing | Missing | 39.2 | 24.2 |
| 2026-11-02 | Missing | Missing | 42.7 | 25 |
Weighted yes responses about the preceding two weeks. The chart breaks at the October 19, 2026 wording change. The national estimate covers all in-scope sectors.
Source · U.S. Census Bureau Business Trends and Outlook Survey; Euphemios analysis · Sector and national core response estimates, November 19, 2026 release · As of
The older survey question asked about use in producing goods or services and cannot be compared with the broader series used here. The broader question began with the November 17, 2025 collection. Census then changed the terminology with the October 19, 2026 collection, retaining the scope of any business function. The chart breaks at the later wording change because its effect on responses is unknown. Before that wording change, the last estimates were 40.5% for Finance and Insurance and 24.6% nationally.
What we cannot measure yet. The survey is self-reported, with sampling and response uncertainty. The broad current-use question leaves the share of work performed with AI, the specific tools used and use in excluded funds unknown.
Filing counts at the common November cutoff
U.S. Securities and Exchange Commission EDGAR full-text search; Euphemios analysis
| Group | 2025 through November 20 | 2026 through November 20 |
|---|---|---|
| All original 10-Ks | 3,067 of 6,237 (49.17%) | 3,817 of 6,266 (60.92%) |
| Banks | 207 of 347 (59.65%) | 265 of 328 (80.79%) |
| Brokers, dealers and exchanges | 23 of 29 (79.31%) | 26 of 31 (83.87%) |
| Investment advice | 40 of 49 (81.63%) | 55 of 60 (91.67%) |
| Insurance carriers and agents | 116 of 140 (82.86%) | 125 of 139 (89.93%) |
| Other credit and finance | 71 of 113 (62.83%) | 102 of 142 (71.83%) |
| Fintech and payments | 10 of 10 (100.00%) | 9 of 10 (90.00%) |
| Alternative managers | 7 of 7 (100.00%) | 7 of 7 (100.00%) |
| Crypto | 6 of 6 (100.00%) | 6 of 7 (85.71%) |
Each cell gives matches out of original filings and the resulting share.
Methodology and definitions
The annual filing comparison uses EDGAR original 10-K primary documents only and counts each accession once. The numerator is the quoted phrase artificial intelligence. The denominator is all eligible originals in the same filing-date window. The comparison uses dates through November 20 in both years. Industry is assigned using returned SIC and excludes real estate investment trusts. Fintech, alternative managers and crypto use stated issuer lists keyed by CIK. Cohorts may overlap the SIC groups.
The earlier November 20 pull for AI across banking, crypto, government contracting and construction found 3,810 matches in 6,258 original filings. Eight November 20 originals that the SEC accepted after the earlier pull, seven of them phrase matches, account for the net increase to 3,817 of 6,266 in the later pull, with identical bank and crypto counts in both reports.
The Form D comparison joins the SEC's quarterly submissions, offerings and issuers by accession. It counts initial and amended filings in the hedge-fund series. Dollar sums use finite offering amounts on initial notices. The name test recognizes standalone AI or artificial intelligence, ignoring case. The hedge-fund group requires the Hedge Fund subtype of Pooled Investment Fund. The financial-services name proxy uses the banking and financial-services industry choices, with dates through September 30 for each partial year.
Tagged expense and software figures use annual consolidated U.S. dollar facts reported with a comparable standard tag in both fiscal years. Coverage applies to the stated list. The Census comparison uses the weighted current-use Yes estimate in the sector and national workbooks. It begins with the broader question introduced in the November 17, 2025 collection and marks the transition to the Census Bureau's Super Intelligence wording in the October 19, 2026 collection.
Limits
The filing results are bounded by report type, filed dates and how the SEC indexes text and returns issuer classifications. A public company may change how it describes a risk or business even if its operations change little. The private-offering results are as filed and can contain errors or duplicate economic activity across affiliated entities. The name test recognizes words in the legal entity field and does not validate the meaning of any match. A later amendment can change the information about an offering.
The survey and filing series have no common denominator. Sector classification in the survey follows the North American Industry Classification System. Industry groups for the filing search follow the SEC's returned Standard Industrial Classification, and the offering notices use their own group choices. These labels cannot be added together or used to translate one measure into another.
FAQ
What qualifies as a filing mention?
A match is an original annual filing whose primary document appears in EDGAR's quoted-phrase search for artificial intelligence. Amendments, attachments and additional appearances within the same document do not increase the count.
Why do the fintech and crypto comparisons use issuer lists?
Fintech spans several industry codes. The crypto list represents selected public issuers with differing activities. The stated lists let readers inspect who entered each comparison and which filings received matches.
Does a hedge fund with AI in its name use AI?
The name match establishes only that the legal name contains the words or abbreviation covered by the screen. The fund may describe a theme, a brand or a strategy. The notice series does not test which tools the fund uses.
How should I read Form D dollar amounts?
The totals add numeric offering amounts on initial notices that meet the scope. An indefinite amount does not enter the dollar sum. The notice's offering amount and its separate amount-sold field answer different questions.
Do the tagged dollar amounts measure AI budgets?
The expense tag includes communications and technology. The software tags cover acquisition or broader software payments. The public records shown cannot allocate either line to AI.
Who is covered by the business survey?
The comparison uses weighted estimates of current use for in-scope employers in Finance and Insurance and for all in-scope employers nationally. It omits funds, trusts and other out-of-scope groups. The chart notes the separate question wordings used in the reported collections.
Company and network names are trademarks of their owners. They are mentioned as facts about public filings and data, not as a relationship or an endorsement.
Sources
- Data as ofU.S. Securities and Exchange Commission EDGAR full-text search; Euphemios analysisRun Original 10-K filings by filing date, January 2019 through November 20, 2026
- Data as ofU.S. Securities and Exchange Commission Form D Data Sets; Euphemios analysisRun Quarterly Form D packages from 2019 through the third quarter of 2026
- Data as ofU.S. Securities and Exchange Commission EDGAR XBRL Company Facts; Euphemios analysisRun Annual consolidated dollar facts for fiscal 2024 and 2025, available through November 20, 2026
- Data as ofU.S. Census Bureau Business Trends and Outlook Survey; Euphemios analysisRun Sector and national core response estimates, November 19, 2026 release