The S&P 500 is the closest thing the American economy has to a boardroom consensus. So it is worth reading what those 500 companies, writing under legal liability in their annual 10-K reports, now say about artificial intelligence, and where in the document they choose to say it.
Via News read every S&P 500 annual report back to 2019. In the 2019 filing season, 100 of the 500 mentioned AI at all. In the most recent season, 446 did — roughly nine in ten. The phrase “generative AI,” which no S&P 500 company used before 2023, appears in 164 of them today.
| Filing year | S&P 500 naming AI | Naming “generative AI” |
|---|---|---|
| 2019 | 100 | 0 |
| 2020 | 114 | 0 |
| 2021 | 139 | 0 |
| 2022 | 162 | 0 |
| 2023 | 196 | 21 |
| 2024 | 371 | 97 |
| 2025 | 446 | 164 |
That AI is now nearly universal in the S&P 500 is, by itself, unsurprising. The more revealing question is where in the report companies put it. A 10-K has a structure with meaning: Item 1 (Business) is where a company describes what it does and how it intends to win; Item 1A (Risk Factors) is where it warns shareholders what could go wrong. The same words carry a very different message depending on which section they sit in.
From a plan to a peril
When we sort each company’s use of “generative AI” by the section it appears in most, a shift appears in the space of a single filing cycle. In 2023, generative AI was described more often as strategy — what the company was building — than as risk. By 2024 that had flipped, and hard: it became something companies felt obliged to file under threats to the business, and it has stayed there.
| Where S&P 500 firms name “generative AI” | 2023 | 2024 | 2025 |
|---|---|---|---|
| Risk Factors (what could go wrong) | 9 | 61 | 105 |
| Business (strategy) | 12 | 35 | 50 |
| Management Discussion (results) | 0 | 0 | 1 |
Two details sharpen the picture. The first is the flip itself: in 2023, the leading home for “generative AI” was the Business section, where a company talks up its plans. One year later, Risk Factors had nearly doubled the Business count, and by 2025 it led by more than two to one. The technology did not become less promising; it became more consequential, and consequences are what the risk section exists to disclose.
The second is what is almost entirely absent. Management’s Discussion and Analysis — the section where a company explains what actually moved its revenue and costs — barely mentions generative AI at all: zero, zero, and one across the three years. Companies are prepared to describe AI as a strategy and to warn about it as a risk, but almost none yet attribute their financial results to it. The disclosure has run ahead of the earnings.
What we are and aren’t claiming
Two honest limits. The section split is assigned by where each company mentions the phrase most; a filing that names it once in each place is classified by the majority, not by nuance. And the panel is the current S&P 500 read backward, so membership changes over the years are not adjusted for. Neither caveat touches the shape: adoption is near-universal, and the framing moved from plan to peril.
The question underneath
Read together, the filings say something the companies would not put this plainly: AI is now material enough to be both a strategy worth advertising and a risk worth disclosing, and the single thing companies fear most about it — as a companion analysis of the same filings shows — is being out-competed by someone else’s. That is a market reorganizing itself around a technology it treats as decisive. Which leaves the question these annual reports are not written to answer: when a force this decisive reshapes who wins, who shares in what it produces? The filings are candid about the stakes. They are silent on the distribution.
