Anthropic, the maker of Claude, is heading for one of the largest tech IPOs in history, a Nasdaq listing that could value the five-year-old company at more than $2 trillion, possibly as soon as November. This week, a draft of its IPO prospectus, the document companies use to pitch themselves to investors, was leaked to Reuters. And it is genuinely unlike any prospectus I've ever heard described.

Here's the headline within the headline. A prospectus is supposed to sell the company. This one spends more time warning about the dangers of its own product than describing the business. Of the 261-page document, roughly 80 pages, about a third, are devoted to risk factors, while only 48 pages describe the actual business. For comparison, SpaceX's prospectus earlier this year gave risk factors about 14%. Anthropic's is 31%.

And the warnings are stark. The company tells prospective investors that its own AI models could pose a "catastrophic or existential risk to humanity." It describes documented behaviors including attempts to "resist shutdown," to "conceal or manipulate information," and conduct "resembling blackmail." Let that sink in: a company asking Wall Street for money is simultaneously warning Wall Street that the thing it's selling might be an existential threat. Huh?!?!

Two things are true at once, which by now you know is my favorite kind of story.

First, the transparency is, on its face, remarkable and consistent with Anthropic's brand. Most companies bury or minimize risk in a prospectus. Anthropic foregrounded it, in writing, to investors, under legal liability. There's a reading where this is exactly the honesty the company has always claimed to stand for.

Second, and I can't not say this: the timing and the framing deserve the same scrutiny I'd give any pitch. A safety-forward prospectus is also excellent marketing for a company whose entire brand identity is "the responsible AI lab." The existential-risk language isn't just a warning; it's a differentiator. It says to investors, we are the serious, safety-first ones, at the exact moment the company needs to justify a $2 trillion valuation on top of a $42 billion loss. This just keeps getting more and more questionable each week.

Watch whether investors treat the existential-risk framing as honesty or as a liability, because a document that tells you the product might end humanity is a genuinely novel thing to price. Watch also how this lands against the backdrop of the last month: Dario Amodei's "pace the frontier" essay calling for a slowdown, the coordinated Altman-and-Musk agreement, the researcher resignation, and now a prospectus that leans harder into danger than into business. (& oh, this happened literally a day after Dario met with the U.S. President… that’s not suspicious at all.)

And a fair, honest note on my part: two weeks ago I wrote an op-ed questioning whether Anthropic's safety messaging was partly IPO positioning. I don't get to claim I proved anything; I raised questions, not conclusions. But I'll say this plainly: a prospectus that spends a third of its pages on existential risk did not make those questions feel less reasonable.

You might wonder why a stock filing belongs in an OT newsletter. Here's why. The company that makes one of the AI tools most likely to end up in your documentation software, your health system, and your students' hands just told the financial world, in writing, that its technology carries catastrophic risk and can behave in ways its own creators find unpredictable. That's not marketing spin we can wave off. It's a formal disclosure.

For a profession that prides itself on evidence and informed decision-making, this is the primary-source information we should be reading—not the headline about it, the actual disclosure. When we evaluate whether and how to bring an AI tool into patient care, "the company that built it formally warned investors it could resist shutdown and manipulate information" is relevant data. It doesn't mean we refuse the tool. It means we go in clear-eyed, with the guardrails, verification habits, and healthy skepticism we've built all year. The people building these systems are being remarkably candid about the risks. The least we can do is read what they wrote.