The crucial things Anthropic’s jumbo ‘risk factors’ won’t tell you

Depending on the kind of person you are, the “risk factors” section of a company’s initial public offering document is either the first place you look or the last. Anthropic’s will be more readable than most, thanks to its stark warning that the AI lab’s products could extinguish humanity. But some more salient information will almost certainly be absent.

The practice of itemising risks to a new stock in its pre-IPO filing, and thereafter in annual reports, is a classic example of good intentions gone awry. Such disclosures have been mandatory in the US since the 1960s. But through a combination of regulatory diktat and fear of lawsuits, they have multiplied. Microsoft in 1986 offered fewer than two pages. SpaceX this year stretched to 38. Uber, in 2018, filled 50.

Anthropic might be the first to warn of human extinction as a potential input into a $2tn valuation, but weird warnings are not unknown. Lab testing firm Charles River advises of the risk of transmitting “new pathogens”. President Donald Trump’s media group TMTG warned in 2024 that the company could be “significantly disadvantaged” were its founder to wind up incarcerated. Nuclear energy firms, naturally, have to nod to the unthinkable.

Is obliterating all humans a risk worth elucidating? If it happens, there will be nobody left to sue, so perhaps not. But noting AI’s apocalyptic potential is still helpful, because if lawmakers impose stringent, profit-sapping curbs on Anthropic’s business model because of such existential fears, lawsuits may fly.

The real question for those considering buying Anthropic stock is whether it will make money. On that, the IPO filing is, at best, only moderately useful. Anthropic, like all companies, is likely to refrain from sharing its executives’ forecasts of revenue and profit. Instead, it will report backwards-looking figures, which are next to useless because of its rapid rate of change. Last year’s revenue was $4.6bn, the FT has reported; already this year, its run rate in July was $65bn.

This lacuna is the result of a legal quirk: companies enjoy protection from legal liability if they disclose forecasts once they’re public, provided they do so in good faith, but not before. Researchers at Arizona State University, University of North Carolina, the University of Washington and the University of Michigan found that one-third of listing companies tackled this problem by giving verbal guidance in their pre-IPO “roadshow”, but even they mostly stick to broad strokes on steady-state profit margins.

Other insights investors might want but are unlikely to get: who are the two customers that made up a quarter of Anthropic’s revenue last year? And if there really is a risk of AI apocalypse, how risky is it exactly? Anthropic could report its view of P(doom), the mathematical probability of mass destruction. It could even provide a table showing what happens to future earnings if the number goes up or down — as banks often do regarding interest rates.

Instead, expect Anthropic to provide the worst of all worlds: fuzzy estimates of its “total addressable market”. It may even try to outdo rival SpaceX, which touted $28.5tn in its IPO filing — an astronomical 600 times the revenue analysts expect Elon Musk’s rocket maker to generate this year. The real risk factor is the likelihood that investors are too easily impressed and end up dramatically disappointed.

john.foley@ft.com

Lex sends a newsletter every Wednesday with additional insights, the pick of the week’s columns and links to what we’re reading. Premium subscribers to the FT can sign up