Pay No Attention to the Man Behind the Curtain: 85% of the Fortune 500 Talk About AI in Their Filings. 11% Actually Run It. [2026]
The Wizard is enormous. His voice fills the hall. Smoke, flame, a floating green head, a roar that shakes the floor. Dorothy and her companions are terrified. Then the dog pulls back a curtain and there is a small man at a console, working the levers, speaking into a microphone.
“Pay no attention to the man behind the curtain.”
The Fortune 500’s AI story in 2026 has a curtain. It is the annual 10-K filing, the earnings call, the keynote. Behind it is a much smaller operation than the voice suggests — and for the first time, the data to see the man at the console is public.
Act One: The Curtain
An analysis of Fortune 500 annual filings found that 85% of companies now mention AI in their 10-K, up from 29% in 2022. Total AI mentions across all filings grew 601% in three years. The tone shifted too: in 2022, 30% of companies discussing AI framed it purely as an opportunity. By 2025 that fell to 4%, with 78% acknowledging both risks and benefits — which is what a company writes when its lawyers have reviewed the paragraph.
Orgvue’s 2026 survey of 1,163 senior decision-makers at large organizations put the investment figure at 92%. Almost everyone has spent money. Microsoft and IDC report 85% of the Fortune 500 “using AI in some capacity.”
That is the voice in the hall. Universal disclosure, universal investment, a 601% increase in the volume of the roar.
Act Two: The Man at the Console
Now the curtain.
Researchers built a five-point rubric for AI adoption and applied it to every S&P 500 company using its own public disclosures. A score of 5 means production-scale deployment integrated into core operations. In 2025, 11% of the S&P 500 scored a 5. Another 10% scored a 4. Eighteen percent had no mention of AI adoption at all. And of the companies scoring 4 or 5, two-thirds were technology-sector firms — which means outside of tech, the number is considerably lower than 11%.
Orgvue’s survey asked the decision-makers what happened to the money. 78% said their AI projects had either failed outright (35%) or remained stuck in pilot (43%). A third said they still do not understand how to make AI work. Nearly half have committed to upskilling programs specifically because they lack the talent to deploy what they bought. Across the broader enterprise data, only 25% of AI initiatives deliver expected ROI and only 16% reach enterprise-wide scale.
Then the two receipts that matter most, because they come from the companies’ own regulated behavior rather than their statements.
The restructurings. Orgvue analyzed the 10-Ks of 475 Fortune 500 companies. 52% reported an organizational restructure last year. Collectively they recorded $49.4 billion in severance. And AI or automation was referenced in less than 10% of those restructures. Nearly three-quarters were driven by cost reduction or organizational simplification. The “AI is replacing workers” narrative is not what the companies themselves tell the SEC when they explain a layoff. They say “cost.”
The hiring. Orgvue’s separate analysis of current Fortune 100 job postings found only 11% mention AI at all, and just 6% name a specific platform or tool. Technology firms are highest at 26%. If AI were transforming the operations of America’s largest companies, they would be hiring for it. Nine in ten postings do not ask.
| The Curtain | The Man Behind It |
|---|---|
| 85% of Fortune 500 mention AI in 10-K filings | 11% of S&P 500 at production-scale adoption |
| 92% of large organizations have invested | 78% of projects failed or stuck in pilot |
| AI mentions in filings up 601% since 2022 | AI cited in under 10% of actual restructurings |
| “Doing more with less” on every earnings call | 11% of Fortune 100 job postings mention AI; 6% name a tool |
The gap between the left column and the right column is not a rounding error. It is the distance between the voice in the hall and the man at the microphone.
Why the Gap Exists (And Why It Is Rational)
The Wizard is not a fraud in the usual sense. He is a man who arrived in a balloon and was mistaken for something larger, and it was easier to keep working the levers than to correct the record. The Fortune 500 AI gap has the same structure.
A 10-K that does not mention AI in 2026 invites an analyst question about why not. A 10-K that mentions it, with appropriately hedged risk language, does not. Mentioning AI is nearly free and not mentioning it has a cost. Every company converges on mentioning it. That is the 85%.
Running AI at production scale is not free. It requires the data infrastructure, the validation layer, the evaluation framework, the change management, and the talent — the whole pipe under the model — and most organizations built the demo before they built the pipe. The research on why adoption is slow is consistent: even frontier models cannot match average worker performance across most task sets, and the economics of building your own are unattractive for nearly everyone. So the rational path is to say it loudly and deploy it slowly. That is the 11%.
The severance data closes the loop. Companies are cutting headcount and telling the SEC it is about cost. If AI were delivering the productivity the filings imply, it would be the stated reason. It is not, because it is not.
Act Three: What Dorothy Does With the Information
Dorothy does not storm out. She asks the man for what he can actually provide — which turns out to be a balloon ride and some perspective. The people who read the Fortune 500 gap correctly get something similar.
If you are the person who can tell the filing from the floor, you have a job. The organization that spent the money, stalled in pilot, and cannot explain why is looking for exactly one person: the one who can walk in, count the seats paid versus the seats active, and put the gap in dollars on a slide. That is the AI Enablement Lead, and this piece is the receipt stack for the interview.
If you are the person who has to sign the filing, the gap is a governance problem. Material statements about AI capability in an SEC document that operational data does not support is the kind of thing that eventually gets a question from a regulator or a plaintiff. The AI Compliance Manager seat exists because someone has to inventory what the company actually runs before someone else writes down what it claims.
If you are in a job search, stop reading the curtain. A company’s AI keynote tells you nothing about whether the role you are considering will involve AI. Its job postings do — and 89% of them at the Fortune 100 do not mention it. The skills the postings actually ask for are the skills the floor actually needs. Read the posting, not the press release.
And if you are wondering when the curtain comes down: Toto does not pull it. The CFO does, in the quarter the AI line item is asked to defend itself against the productivity it was supposed to deliver. That quarter has not arrived at most companies. The severance data says the productivity has not either.
I spent a decade in audit. The job is not catching liars. The job is noticing when the story a company tells and the numbers a company produces have stopped describing the same thing — and the gap here is the widest I have seen between a narrative and a ledger since the early 2000s.
The filings say 85%. The production data says 11%. The severance filings say the layoffs are about cost, not automation. The job postings say nine in ten roles do not need the skill. Every one of those is a company’s own regulated disclosure. Nobody is accusing anyone of anything. The companies are simply saying two different things in two different documents, and one of those documents is designed for the hall and the other for the ledger.
This is not a reason to be cynical about AI. The 11% is real, it is mostly in tech and finance, and it is doing real work. It is a reason to be precise about which companies are in the 11% and which are working the levers — because the career decisions, the investment decisions, and the governance decisions all depend on knowing which room you are standing in.
The Wizard gave Dorothy good advice once the curtain was down. He was not useless. He was just not what the voice claimed. Read the postings. Read the severance line. Then decide how much attention to pay to the man behind the curtain.
Sources
Pablo Rios, “Fortune 500 AI Disclosure Analysis: How America’s Largest Companies Talk About AI in SEC Filings,” April 2026, dataset on GitHub (85% mention AI in 10-K, up from 29% in 2022; 601% growth in mentions; opportunity-only framing 30% → 4%) · “AI Adoption in S&P 500 Firms,” arXiv 2607.08920, July 2026 (five-point adoption rubric; 11% scored 5; 21% scored 4 or 5; 18% no mention; 67% of high scorers in technology sector; Acemoglu & Lensman slow-convex adoption model) · Orgvue, “92% of organizations have invested in AI but 78% say projects have either stalled or failed,” May 12, 2026 (1,163 senior decision-makers; 35% failed, 43% stuck in pilot; 32% do not understand how to make AI work; 49% upskilling for lack of talent) · Orgvue, “Myths and reality: AI workforce dynamics across the Fortune 500,” June 2026 (475 companies’ 10-Ks; 52% restructured; $49.4–49.7B severance; AI referenced in under 10% of restructures; 73% cost/simplification driven) · Orgvue, “The AI hype check: Fortune companies are not hiring AI-ready workers,” October 2025 (Fortune 100 postings: 11% mention AI, 6% name a tool, 26% in tech sector) · Microsoft/IDC via Fortune AIQ 50 dataset, February 2026 (85% using AI in some capacity; maturity tiers) · Enterprise AI agent statistics compilation, September 2026 (25% deliver expected ROI; 16% reach enterprise-wide scale).
