When a public company lists artificial intelligence among the things that could go wrong, it has to say what, specifically, it is afraid of. Via News read the Risk Factors section of every S&P 500 annual report that names “generative AI” — 221 of them — and sorted the fear by type. The result overturns the usual assumption about what worries corporate America.
It is not the hallucinations. It is not the robots. The single most common fear, named in 44% of these filings, is being out-competed — that a rival will adopt AI faster or better and leave the company behind. The technology breaking down (accuracy, hallucination) ranks near the bottom, at 19%.
| What they fear about generative AI | Share of filings |
|---|---|
| Being out-competed (a rival moves faster) | 44% |
| Regulation & compliance | 37% |
| Reputation & trust | 34% |
| Legal, IP & copyright | 30% |
| Privacy & data | 30% |
| Ethics & bias | 29% |
| Security & misuse (fraud, deepfakes) | 28% |
| Accuracy & hallucination | 19% |
| Cost of investment | 17% |
| Talent & workforce | 5% |
Afraid of each other
The language is remarkably consistent, and it comes from companies with nothing to do with building AI. Invitation Homes, which rents out single-family houses, warns that its “competitors or other third parties may incorporate generative AI into their information systems and operations more quickly or more successfully than us, which could impair our ability to compete effectively.” Vistra, a power generator, files nearly the same sentence: “Competitors may also incorporate emerging technology like generative AI into their businesses, services, and products more quickly or more successfully than we do.” Oracle simply states that “AI technology and services are highly competitive and rapidly evolving.”
Read across the market, the fear is not of the machine. It is of the neighbour with the better machine. A landlord and a utility do not worry that generative AI will malfunction; they worry that whoever adopts it first will win, and that standing still is the real risk. That is what a technology looks like once it is understood to be decisive: the danger is no longer whether it works, but whether you moved fast enough.
The governance cluster
Behind competition sits a dense band of governance worries, each named by roughly a third of filings: regulation (37%), reputation (34%), legal and intellectual-property exposure (30%), privacy (30%), and bias (29%). Together they describe a technology companies feel they must adopt to survive and cannot yet fully control — powerful enough to be a competitive necessity, unsettled enough to be a legal and reputational minefield. Notably, only 17% frame the cost of the investment as the risk. Spending the money is assumed; the fear is what happens around the spending.
What we are and aren’t claiming
These are counts of language, categorised by keyword within each company’s risk section; 142 of the 221 filings used specific enough wording to classify, and categories overlap because a single filing usually names several fears. The exercise measures what companies choose to disclose as risk, not the underlying reality of their exposure. But disclosure is a choice made under legal liability, and the choice is clear: the thing to fear about AI is falling behind in it.
The line this draws
Put the two facts together. Companies now describe AI as a force decisive enough that the chief danger is being out-run by it, and the race rewards whoever owns and deploys the technology best. That is precisely a mechanism for concentrating the gains — toward the fastest adopters and the owners of the capital that funds them. The filings are unusually honest about the stakes of the race. They say nothing about who, beyond the winners, ends up sharing in what it produces. That silence is the subject of a larger argument; the filings simply mark where it begins.
