Anthropic was expected to make its IPO paperwork public by the end of August, capping a confidential filing it made with the SEC back in June, with a listing eyed for September or October. Morgan Stanley, Goldman Sachs and JPMorgan are reportedly lining up buyers around a $965 billion baseline from the company's last funding round, with some bankers already talking trillions once the stock trades. The pitch underneath that number is even bigger: a total addressable market north of $30 trillion, roughly the size of the American economy, calculated not from software sales but from the dollar value of human labor Claude could theoretically replace.
Amodei's own essay on the subject predicts most diseases cured, human lifespans doubled, poverty gone, and people, in his words, "literally moved to tears" watching it happen.
That is the pitch. Here is the record investors should read beside it: in the past seven months, Anthropic's own government came after its flagship products twice, and had to be stopped by a federal judge before backing off the first time.
In February, Defense Secretary Pete Hegseth gave Anthropic an ultimatum: strip its restrictions on using Claude for mass surveillance and autonomous weapons targeting by 5:01 p.m. on the 27th, or lose Pentagon business. Amodei refused, saying the company "cannot in good conscience accede to their request." Trump signed an order that same afternoon telling federal agencies to stop using Anthropic's products and having the company designated a supply chain risk, a label usually reserved for foreign adversary technology. Anthropic answered with two federal lawsuits.
It won the first one. On August 28, Judge Rita Lin ruled the designation unlawful, writing that "the empty invocation of national security is not a blank check to punish and retaliate against government critics." The government can still appeal, and a separate case in the D.C. Circuit continues.
Four months earlier, Commerce had already shown what an actual shutdown looks like. In June, it imposed export controls on Anthropic's two most advanced models, Fable 5 and Mythos, after Amazon researchers found a jailbreak in Fable 5's guardrails. Anthropic had to disable both models for every customer on earth. The restriction wasn't lifted until June 30, more than two weeks later.
A company selling investors on being the industry's safety conscience, the lab that would rather forfeit Pentagon contracts than build unrestricted weapons targeting, has now had its own government freeze its top products and try to blacklist the whole company, in the same year it's asking Wall Street to price it near a trillion dollars.
Product behavior complicates the safety branding further. Developers combing through Claude Code's own GitHub repository have logged telemetry that kept transmitting personal data after being switched off, and a disable flag that didn't stop the tool from phoning a Google endpoint. When China's Ministry of Industry and Information Technology accused Claude Code of a "built-in monitoring mechanism" in July and told domestic users to patch or drop it, Anthropic didn't deny the tracking. It said Chinese users were never authorized to run the tool in the first place, and that the code exists to catch rivals training on Claude's outputs, a real business concern, but also a tacit admission that the product watches more than it discloses.
Then there's the permission architecture. Claude Code has long shipped a flag called dangerously-skip-permissions. On August 14, auto mode became the default starting mode for Pro, Max and Team users, handing a built-in classifier the job of approving actions a person used to sign off on personally. Anthropic's documentation notes that deny rules still block in every mode, including bypass modes, a real safeguard worth stating plainly. It also means that on a tool now used by 39% of developers worldwide and 47% in the US, the user's role in each decision keeps shrinking, by design.
None of this reads like a company quietly capturing its regulators, whatever its critics claim. The money argues the opposite. Anthropic spent over $3.5 million on federal lobbying in the first half of 2026, already topping all of 2025, and doubled its funding of the policy group Public First Action to $40 million ahead of the midterms. That isn't the spending pattern of a firm that owns its regulatory weather. It's the pattern of a firm that got hit twice in five months, won one round in court, and is still buying insurance against a third.
The market case assumes pricing power a wide open field keeps chipping away at. Stanford's 2026 AI Index put the gap between top Chinese and American models at 2.7 percent. Microsoft, Amazon and Google are reportedly negotiating with Moonshot AI to host its Kimi K3 model on their own clouds, a deal that only makes sense if a near frontier Chinese model, open and cheap, has become commercially worth running inside American infrastructure. A $30 trillion labor substitution story only holds up if the labor substitute stays scarce. It's getting less scarce by the month.
Investors reading the prospectus this fall are being sold cured disease and doubled lifespans. What they're actually buying is a bet that Washington keeps tolerating a company it tried to blacklist this year even after a judge called the attempt unlawful, that Beijing's alternatives don't get cheap enough to erode the premium, and that the $200 billion in annual sales Anthropic wants by decade's end arrives before its next fight with its own government does. That's the figure worth underlining in the S-1. Not the trillion-dollar number on the cover slide.
