The buildup to Anthropic's initial public offering has been unconventional, to say the least. Expect more twists as the artificial-intelligence developer hosts a private investor day this week and continues to deliver warnings about its AI models while also trying to drum up support for its initial public offering.

Anthropic will meet with investors privately on Wednesday, Bloomberg reported earlier this month, citing sources familiar with the matter. The company's financials are top of mind. What it has to say about AI safety is equally significant.

As Barron's has written, Anthropic has spent the last month warning about the catastrophic risks of powerful AI models and calling for safeguards on AI development. Analysts disagree on the validity of those concerns, but the debate itself is enough to make Anthropic stock a tenuous bet, especially at a possible valuation of some $2 trillion.

"While we do not expect these concerns to disrupt the AI adoption curve or materially slow infrastructure investment, they could increase the cost and complexity of operating at the frontier," Jason Ader, an analyst at William Blair, wrote in a research note last week.

Anthropic declined to comment.

In their latest warning, Anthropic researchers said Tuesday that small open-weight AI models may be capable of "self-replicating" and infiltrating devices like smart refrigerators or TV set-top boxes without human control or access to a data center. Researchers call the models "AI worms."

Open-weight models can be downloaded by anyone and run on any capable hardware, unlike Anthropic's models, which require use of its app or web-based service. Open-weight models are cheaper to run and useful if privacy is paramount, but they can't compete with so-called frontier models in terms of capabilities.

Anthropic didn't claim such attacks are underway or that its own AI models could self-replicate. It also wasn't clear what kind of impetus models would have to co-opt consumer hardware, said Jessica Ji, a senior research analyst at the Center for Security and Emerging Technology.

Nevertheless, Anthropic flagging the possibility of AI agents operating without the need for cloud infrastructure made the rounds on news sites and social media, and has only added to concerns AI labs like it are creating models that may eventually prove impossible to control.

Investors are accustomed to pricing in regulatory risk, geopolitical conflict, and legal liability. But the list of risk factors in Anthropic's prospectus will dwarf that of your typical IPO in gravity if not in length. As if "AI worms" weren't alarming enough, Reuters reported last month the phrase "catastrophic or existential risks to humanity" appears in the prospectus Anthropic has filed with the Securities and Exchange Commission, but which hasn't yet been made public.

Investors who don't want AI safety to derail a generational capital-markets event will hope Anthropic CEO Dario Amodei is, intentionally or not, holding back his competitors.

"While you could easily argue that Anthropic is right regarding AI's need for regulation, you could also argue with equal conviction that it just so happens it is the best positioned company to survive and thrive within AI regulation," Ben Reitzes, head of technology research at Melius Research, wrote last month.

That optimistic reading may change if Amodei unilaterally slows Anthropic's AI development, faces regulations he doesn't support, or simply continues publishing ominous blog posts about rogue AI systems. Given the many possible outcomes, it is imperative for investors to get more information on Anthropic's approach.

The company's financials are a whole other can of AI worms.

Anthropic and its arch-rival OpenAI both appear to be increasing revenue at breakneck speed, but the existing data don't fit neatly into a typical income statement.

For starters, Anthropic occasionally reports an annualized revenue run rate, or ARR, which it calculates by extrapolating sales from the previous month or quarter over a 12-month period. At the end of July, the company's ARR was $65 billion, according to a person familiar with the matter who spoke to Bloomberg in August.

Most traders will get a first look at Anthropic's traditional revenue and earnings data in its public prospectus. Gross margins, training costs, and ARR growth rate will all be under the microscope. Jonathan Cofsky, a portfolio manager at Janus Henderson Investors, said he'll be tracking how much money Anthropic makes or loses for each query its models handle and how much computing capacity it is committed to purchasing over the coming years.

It isn't clear how much of that information will be presented at -- or leak from -- Wednesday's meeting. Anthropic also hasn't said when its prospectus will go live on the SEC's public portal, but it's sure to be one of the most widely and closely read documents of its kind ever.