On October 26, Anthropic announced that it had filed a registration statement on Form S-1 with the SEC for a proposed initial public offering, with plans to list on Nasdaq under the symbol "ANT." The filing is preliminary. The share count and price range are still blank.
The risk factors are worth reading closely, because they show what happens to a safety story when it has to go into a prospectus.
The claim, with its limits attached
Until now, AI safety has mostly been told in blog posts, system cards and keynotes. In a prospectus, the same story has to be written as the ways it might not hold. Anthropic's filing does that in plain language. Of its technical safeguards against misuse of its models to develop chemical, biological or other dangerous weapons, it says they are "incomplete, may be outdated in light of evolving threat models, have been and may in the future again be circumvented, or may fail to operate as intended." On release decisions, it says: "We often make judgment calls with respect to when we determine a model is safe enough to release and which capabilities to develop or restrict." And it adds: "There can be no assurance, however, that the safeguards or access regimes we adopt will prove adequate, that we will correctly identify which capabilities require restriction, or that constraints we impose will not be circumvented."
Anthropic also acknowledges what a filing like this does. Because its disclosures describe risks and model behaviors in detail, they "could be cited by regulators, litigants, or others in proceedings or claims against us."
I want to be fair here. Risk factors are cautious by design, and every company's lawyers write them that way. But these sentences are specific. Some describe limits that have already shown up, not only hypothetical ones. The filing also describes some of the response: "a program under which independent evaluators work alongside our own risk and safety teams with broad internal access, subject to legal, confidentiality, and security limits." That is not a gotcha. It is a company moving from "trust us" toward "check us." That is a step toward proof.
Governance has to be legible too
The same discipline applies to who can say no. The filing says Anthropic has "a novel capital and corporate governance structure that has no direct public company precedent." A Long-Term Benefit Trust generally has the authority to elect and remove the directors who will make up a majority of the board, and the co-founders' LLC will hold a share of Class F stock carrying 50.1% of the voting power on matters put to stockholders generally. Anthropic lists the structure under its governance risks. A governance structure is a trust claim too, and it needs the same evidence.
From claimed to proven
In Built for Trust, I describe the Trust Maturity Model™ as five levels: Trust Theater, Trust Awareness, Trust Coordination, Trust Integration and Trust Capability. At Trust Theater, the claim is made without a system behind it. At Trust Capability, the proof exists before anyone asks for it.
A risk factor is a trust claim with the theater removed. It has to say what the claim covers, where it has failed, and who is accountable. Most organizations have never had to write their trust claims that way. Increasingly, someone will ask them to. Speak less. Govern more.
A test for your board
Take your three most important trust claims about AI, data, security or service, and rewrite each one as a risk-factor sentence, the way Anthropic's filing does. Then ask:
Could we say where this claim has already failed, and not only where it might?
Who owns the evidence, and who outside the team can check it?
If this sentence were read back to us in a dispute, would our records support it?
Which of our trust claims would survive being written as a risk factor? The ones that would not are the ones to fix, or to say less about, before someone else reads them aloud.
Where does your organization stand? Take the Trust Maturity Diagnostic™ to see which level of the Trust Maturity Model describes you today.
Built for Trust arrives July 27, 2027.
