Artificial intelligence went from a rounding error to the majority of American annual reports in eleven filing seasons
Percentage of 10-K annual reports filed with the SEC between January and June that contain each exact phrase at least once.
Source: U.S. Securities and Exchange Commission, EDGAR full-text search. Retrieved August 24, 2026. Analysis by DFX Intelligence.
What the data shows
Every American public company files an annual report, and the overwhelming majority file it between January and June. That makes the 10-K corpus something unusual: a near-census of what large companies believe they are obliged to tell investors, taken at the same point every year, with a denominator that barely moves. Counting a phrase across it is not a survey with a response rate. It is closer to a complete count.
In the 2015 season, 27 of 7,328 annual reports contained the phrase “artificial intelligence”. In 2026, 3,639 of 6,246 did. The rate went from 0.4% to 58.3%. Over the same eleven years the number of annual reports filed in that window fell by 15%, which removes the one obvious alternative explanation: there are not more filings, there are more filings that talk about this.
The curve has one clear inflection. Between 2023 and 2024 the rate went from 15.4% to 32.5%, a single-season gain of 17 percentage points, larger than the entire first nine years of the series combined.
Companies have adopted the category and avoided the technology
Share of 2026 annual reports containing each exact phrase.
- artificial intelligence+58.3%3,639 filings
- machine learning+30.8%1,924 filings
- generative AI+9.4%588 filings
- large language model+0.9%59 filings
- AGI+0.1%4 filings
Source: U.S. Securities and Exchange Commission, EDGAR full-text search. Retrieved August 24, 2026. Analysis by DFX Intelligence.
“Generative AI” reached 9.4% and “large language model” 0.9%, against 58.3% for the category as a whole. Artificial general intelligence appears in 4 filings out of 6,246.
There is a plain reading of that gap and it is probably the right one. A disclosure lawyer writing a risk factor wants language broad enough to cover whatever happens and specific enough to have warned about it. “Artificial intelligence” does that. “Large language model” commits the company to a claim about a particular technology, which is a liability rather than a shield. What this measures is therefore not adoption. It is the point at which a subject became something a company could not leave out.
What this does not tell you
A mention is one occurrence anywhere in a document that routinely runs past a hundred pages. A company describing artificial intelligence as an existential threat to its business model and a company describing it as its core strategy both count once, and this dataset cannot distinguish them. Nor can it tell you whether the company does anything at all with the technology. The honest description of this series is that it measures what American public companies now feel required to address in writing, and on that narrow question it is close to a complete count.
The full data
| Season | 10-Ks filed | artificial intelligence | machine learning | generative AI | large language model | AGI |
|---|---|---|---|---|---|---|
| 2015 | 7,328 | 0.4% | 0.5% | 0.0% | 0.0% | 0.0% |
| 2016 | 6,941 | 0.6% | 0.8% | 0.0% | 0.0% | 0.0% |
| 2017 | 6,624 | 1.3% | 1.8% | 0.0% | 0.0% | 0.0% |
| 2018 | 6,617 | 3.4% | 2.9% | 0.0% | 0.0% | 0.0% |
| 2019 | 6,317 | 5.5% | 4.7% | 0.0% | 0.0% | 0.0% |
| 2020 | 6,331 | 7.1% | 5.8% | 0.0% | 0.0% | 0.0% |
| 2021 | 6,818 | 10.3% | 7.8% | 0.0% | 0.0% | 0.0% |
| 2022 | 7,334 | 13.8% | 10.5% | 0.0% | 0.0% | 0.0% |
| 2023 | 6,981 | 15.4% | 10.9% | 0.2% | 0.0% | 0.0% |
| 2024 | 6,445 | 32.5% | 16.4% | 4.8% | 0.4% | 0.0% |
| 2025 | 6,252 | 46.9% | 23.1% | 7.9% | 0.6% | 0.0% |
| 2026 | 6,246 | 58.3% | 30.8% | 9.4% | 0.9% | 0.1% |
Methodology and sources
Share of SEC annual reports mentioning artificial intelligence
- Produced by
- U.S. Securities and Exchange Commission, EDGAR full-text searchTIER A
- Retrieved
- August 24, 2026 at 10:22 PM UTC
- Observations
- 12 rows from 7 upstream series or queries
For the January-to-June filing season of each year from 2015 to 2026, the number of 10-K filings whose full text contains the exact phrase, divided by the number of 10-K filings in the same window retrieved by the same query with the phrase removed. Counts are of DOCUMENTS containing the phrase at least once, not of mentions.
- A mention is not adoption. A phrase appearing in a risk factor and the same phrase appearing in a strategy section both count once.
- EDGAR full-text search covers filings from 2001 onward and matches exact phrases.
- Counts are of filings, not companies: a company filing an amended annual report in the same window is counted twice.
- Each observation covers 1 January to 30 June, the window in which the large majority of American annual reports are filed. Holding the window fixed keeps the population of filers comparable from year to year.
- completeness12 of 12 annual filing seasons
- same_seasonwindows end 2015-06 and 2026-06
- denominator_stableannual reports in the January-June window range 6,246 to 7,334 across 12 years (17% spread), so a rising rate is not an artefact of filing volume
- phrase_containmentthe narrower phrase never outcounts the phrase it contains
- exact_totalsevery total was returned by EDGAR with relation 'eq'; capped totals raise rather than publish
- same_form_typenumerator and denominator are both counts of 10-K filings over identical date windows
Source: U.S. Securities and Exchange Commission, EDGAR full-text search. Retrieved August 24, 2026. Analysis by DFX Intelligence.
Observatory data may be reused with attribution to DFX Intelligence and a link to this page. Underlying government data is in the public domain. Our editorial standards and corrections policy explain how findings are checked and how errors are fixed.