SEC 10-K

Anyone who buys shares in a company gives it money and becomes a co-owner. So that this transaction doesn’t rest on mere guesswork, US law requires a comprehensive report once a year. This report is called the 10-K, because the corresponding form is numbered that way. It is filed with the SEC, the American securities regulator that oversees stock trading. A 10-K is not a promotional brochure but often a document several hundred pages long covering revenue, profit, debt, risks, and legal disputes. Every company whose shares are publicly traded in the US must submit it — Apple and Nvidia just as much as small firms hardly anyone has heard of.

Why investors trust the 10-K more than the press release

Companies like to talk about themselves in flattering terms. A press release is allowed to emphasize what went well and omit what went poorly. In the 10-K it’s different: the information is legally required, and false statements are a criminal offense. The CEO and CFO must personally sign to confirm that the report is accurate. This signature requirement was introduced after major accounting scandals around the year 2001.

On top of that, the figures are checked by an independent auditor. This is an external firm that reviews the books and puts its judgment in writing in the report. An auditor who raises doubts immediately triggers nervousness on the stock market.

The “Risk Factors” section is especially widely read. There, a company itself must write down what could go wrong. When chipmakers began listing export restrictions to China as a risk in 2023, that was an important signal for analysts. Such sentences rarely appear in a press release.

The document’s fixed blueprint

A 10-K always follows the same structure, no matter which company files it. Part one describes the business: what is sold, to whom, in which countries. This is followed by the risk factors and disclosures on ongoing legal proceedings. Part two contains the core content for investors, namely the financial statements with the income statement, balance sheet, and cash flows.

Between the tables sits a chapter in which management explains the figures in its own words. It’s called MD&A, short for “Management’s Discussion and Analysis.” There, the company explains, for example, why revenue rose by twelve percent. These explanations are often more revealing than the bare numbers themselves.

Everything is filed digitally in a public SEC database called EDGAR. The reports are stored there in machine-readable form, meaning: programs can read the figures directly without a human having to type them in. A common mistake is confusing the 10-K with the glossy “Annual Report” that companies send to shareholders. That one is designed and tends to gloss things over, while the 10-K is the sober mandatory version. There’s also a shorter quarterly sibling, the 10-Q.

From filing cabinet to fodder for language models

10-K reports constantly turn up in financial news. When a headline says a corporation has “acknowledged for the first time in its annual report” that a business segment is shrinking, that usually comes from a 10-K. Journalists specifically search the documents for changes compared to the previous year, because a newly inserted risk sentence reveals a lot.

For AI applications, these reports are a stroke of luck. They are free, well structured, linguistically consistent, and go back decades. That’s why they are popular for testing language models: can a model pick out the right number from 300 pages? Several benchmark tasks for financial AI consist of exactly such questions.

In practice, you encounter this in tools that summarize reports. Fund managers have key statements from hundreds of 10-Ks output to them as bullet points. This saves time but carries a risk: if the model invents a number, it barely stands out in the running text. Reputable providers therefore link every statement back to the original page in the document.