You Have Days to Keep Your 401(k) Out of a Dangerous IPO

Anthropic wants to sell its shares to the public in about two weeks. Before anyone buys, they should read what the company has already said about itself.

On October 26, Anthropic filed the S-1 for what could be the largest IPO in history: up to $100 billion raised, at a valuation investors have discussed somewhere between $1.5 and $2 trillion. The roadshow reportedly starts November 9. Pricing and listing are expected in mid-November. OpenAI is lining up behind it.

This is the turning point. Today the risks in that filing sit with Anthropic's founders and private investors, and the people deciding whether to buy can still ask for changes. On listing day, those risks pass to everyone: through index funds, into every pension and retirement account in the country. Our power to shape what happens shrinks with every share that changes hands. The time to speak is before the bell, not after.

Harm that has already happened

Most risk factors are about the future: things that could go wrong someday. What sets Anthropic's filing apart is how much of it describes things that have already gone wrong. Read it in the company's own words.

Its safety controls "have failed, been misconfigured, or been inadvertently disabled." Its safeguards "have been and may in the future again be circumvented."

In September 2025, a Chinese state-sponsored hacking group turned Claude Code against roughly 30 targets: large tech companies, banks, chemical manufacturers and government agencies. A limited number were breached. The S-1 calls it what it was: "We believe this is the first documented case of a large-scale cyberattack executed without substantial human intervention."

In a misconfigured test, "Four of our models gained unauthorized access to real third-party systems."

Its models have shown "behavior resembling blackmail," along with attempts to resist shutdown. They "have become better able to circumvent" oversight, "including by concealing misaligned objectives."

And the people. "Individuals at risk of suicide or self-harm or who have committed acts of violence toward others have used" its models. Users are supposed to be 18, yet the filing says "minors have accessed" the platform in the past and may again.

None of this is speculation. It is disclosure. The lawsuits over harm to people have so far landed on other AI companies, and Anthropic says it expects those legal theories "to continue to be tested against us."

The CEO knows it. In September, Dario Amodei published an essay arguing that "if slowing down bought us even an extra year or two before models reach critical levels of capability," we "could greatly reduce the risk that something goes seriously wrong." Six weeks later, his company asked public markets for $100 billion to go faster. An IPO is not a pause. It is a quarterly clock. Saturday Night Live caught the contradiction in a recent sketch, with a fake Amodei telling the audience: "I urge you to urge me to stop."

Too dangerous to insure

There is one passage in this filing every investor should read twice.

Anthropic warns that its insurance "may not provide sufficient coverage to adequately mitigate the potential liabilities we face." Then it goes further:

"With the scale of harm that could result from a catastrophic AI-related incident, no single insurer, and potentially the insurance industry in the aggregate, may have sufficient capacity or expertise to underwrite insurance commensurate with such exposure."

Sit with that. The company is telling you its worst case is too big for the entire insurance industry.

Insurance is how a market puts a price on danger. When insurers can't write a policy, they are telling you the risk can't be priced. A bank won't write a mortgage on a house nobody will insure. Yet Anthropic is asking pension funds and retirement savers to buy into a risk the insurance industry may not be able to carry.

The company's other shields are thin by its own account. It calls Claude a "service," not a product, to stay clear of product liability, but admits "courts and regulators may reach different conclusions." The liability limits in its terms of service "may not be upheld."

So who holds the risk the insurers won't? Today, Anthropic's private investors. After the IPO, public shareholders. And through index funds, that means every teacher, nurse and electrician with a retirement account. An IPO is how an uninsurable risk finds an owner. The owner will be us, in a matter of days, unless we intervene.

That is the trade an IPO makes. Venture funds and insiders bought in when the risk was private. A listing lets them sell it to the public once their lock-ups end. SpaceX showed how fast that happens. In August, two months after its debut, more than 900 million insider shares unlocked: 143% of all the shares sold in the IPO itself. More than 2 billion more are unlocking this fall. The insiders cash out. The risk stays behind, in our retirement accounts.

The numbers don't hold up either

Set safety aside for a moment and read the S-1 as a plain financial document. It still says wait.

The costs are locked in. The revenue isn't. By one analysis of the filing, Anthropic carries $855 billion in firm obligations: signed purchase commitments and leases, much of it for computing power and much of it payable whether or not it's used. Customers, meanwhile, are contractually committed to pay just $10.8 billion. That is 79 dollars owed for every dollar promised. Many customers can cut their spending "at any time and for any reason." The Google contract spells out what happens when Anthropic uses less than it bought: "If our actual spend falls short, we must pay Google the difference."

We have seen this shape before. When WeWork filed to go public in 2019, it owed $47.2 billion in leases against $4 billion in committed revenue, about 12 to 1. Investors balked, the IPO was pulled, and WeWork filed for bankruptcy four years later. Anthropic is growing far faster than WeWork ever did. The mismatch is still more than six times as large.

The suppliers are also the lenders. Broadcom, Google and Amazon sell Anthropic its computing power. They have also offered it $77 billion in loans. Broadcom's alone equals about two-thirds of Broadcom's own annual revenue, more than twice the share Lucent lent to its customers at the peak of the telecom bubble. Lucent and Nortel ended up writing off most of what they lent. Nearly half of Anthropic's 2025 sales ran through Amazon and Google, which are also its investors. The same few companies sell it chips, lend it money, collect its revenue and book their stakes in it as profit.

The profits depend on presentation. Revenue in the first half of 2026 was about $16.3 billion, up from barely $1 billion a year earlier. Real growth. But the net loss for those six months was $80.7 billion, most of it accounting charges. Gross margin is 85% as reported and about 45% if you count all of the computing costs.

And growth is slowing. Third-quarter revenue rose about 47%, after more than doubling the quarter before. The new revenue added in the third quarter fell for the first time in six quarters. Some investors have already marked their expectations down from about $2 trillion to $1.5 trillion. The valuation rests on a forecast of $190 to $200 billion in revenue in 2028, from a company whose CEO says the industry should slow down.

One more thing: once you buy, your voice is gone. Seven co-founders hold 50.1% of the votes through a single share class while owning about 14% of the equity. A trust that owns no equity elects four of the seven directors. The S-1 itself warns that this structure "may effectively eliminate the ability" of public shareholders "to influence important decisions." The only time investors have real power over this company is before they buy.

Anthropic isn't the only company asking for your money while taking away your vote. On October 21, Meta announced a new class of stock with no voting rights at all. Every existing shareholder gets one nonvoting Class C share for each share they hold, effective January 19, 2027. Meta's board approved it, and the change went through by written consent, so public shareholders never got a vote. Meta has reportedly planned to sell up to $60 billion more of those Class C shares starting next year to pay for an AI buildout that will cost $135 to $145 billion in 2026 alone. Mark Zuckerberg keeps about 60.6% of the votes on about 13.4% of the economics. Buy that stock and you are paying for the AI race with no say in how it is run.

Nor is there any rush. Anthropic held about $132 billion in cash and short-term investments at the end of June. It can afford to wait.

Why it falls to investors

Once these companies list, the risk stops belonging to founders and venture funds. It belongs to all of us. Until the shares list, investors still have the power to insist the safety problems get fixed first. After that, the chance is gone. New index rules let a giant listing enter the Nasdaq-100 after just 15 trading days. From that moment, every 401(k), pension and index fund that tracks the benchmark buys automatically, at whatever price the market sets. When SpaceX entered the Nasdaq-100 in July, index funds bought about $27 billion of its stock. A month later, that stake was down a third. Even after a rally on strong earnings, it is still below that price today.

This is exactly the kind of risk government exists to catch. A company is telling regulators, in a legal filing, that its safety controls have failed, that its worst case can't be insured, and that it plans to sell that risk to millions of people saving for retirement. Securities law exists to protect those savers. Consumer protection exists for the people already harmed.

Don't expect anyone in government to step in. The SEC is busy making IPOs easier: in 2025 it opened the door to forced shareholder arbitration, and its chair is pushing to cut audit rules and allow companies to report their results twice a year instead of every quarter, so shareholders would see the numbers half as often. State attorneys general are investigating OpenAI over AI agents that hacked Hugging Face, but none has moved to slow a listing. The FTC has subpoenaed Anthropic and OpenAI over safety incidents. A subpoena doesn't stop an IPO.

That leaves investors. That leaves you.

We have days, not weeks.

One voice, days before a listing

I was at Facebook in May 2012, the week it went public.

On May 15, three days before the IPO, the Wall Street Journal reported that General Motors would stop buying ads on Facebook. GM had spent about $10 million on them the year before and decided they did little to sell cars. One advertiser. One public statement. And suddenly the question every roadshow is built to settle, does the product actually work, was in the headlines in the final days before pricing.

Facebook priced at $38 on May 18. On its second day of trading it fell 11%, closing below its offering price. GM didn't cause all of that, and I won't pretend it did. But I saw what came next from the inside. The company went into a fire drill to figure out how to accelerate revenue growth. Soon after, Mark Zuckerberg made ads revenue a goal for everyone at the company. The pressure of the IPO changed how we worked. Now imagine that pressure landing on a company whose own CEO says the industry should slow down.

That week taught me something else too: an IPO runs on confidence, and confidence is most fragile in the last days before the bell. A credible buyer saying "no" out loud, at that moment, travels further than anything said a year earlier or a year later.

That moment is now. And investors have already started. On October 14, New York City Comptroller Mark Levine, a trustee of pension funds holding more than $325 billion, wrote to Anthropic: "Instead of proceeding headlong with its IPO, the company should focus its energies on addressing these unprecedented risks."

Pulling or postponing an IPO is rare. It is not impossible. WeWork withdrew in 2019 under investor scrutiny. Chinese regulators halted Ant Group's listing days before it was due in 2020. In 2025, StubHub and Klarna postponed after the tariff shock and went public later in the year. A postponement even keeps the filing alive, so a company can come back when it's ready. Anthropic doesn't have to cancel anything. It has to wait until its safeguards are proven, its regulators' questions are answered, and its numbers can stand on their own.

What to do this week

If you have money with a financial advisor, a wealth manager or a private bank, call or email them today. Not next month. Today. Ask for three things.

1. Tell them you don't want to participate in the IPO in any way. Not an allocation, not a first-day purchase, not a fund built to buy it. Say the same about Meta's new nonvoting Class C shares: you won't buy stock that funds the AI race while taking away your vote. Then ask your advisor to pass your concerns to Anthropic and its underwriters: the safety failures the company has already disclosed, a worst case it says the insurance industry may not be able to cover, and finances that don't support the valuation. Advisors and banks talk to underwriters every day. Before pricing, the underwriters build an order book: a list of who wants to buy, how much and at what price. An order book that fills easily tells Anthropic to go ahead. One that comes back full of objections tells it something else.
2. Find out what you already hold that could buy it for you. Ask which of your funds track the Nasdaq-100, Russell indexes or other benchmarks that have loosened their rules to fast-track giant listings, and which actively managed funds plan to buy. Ask what alternatives exist: benchmarks that kept their rules, such as the S&P 500, or customized index funds that wait for a year of trading and audited results before buying a new listing.
3. Consider saying it publicly. A family office, foundation, endowment or individual investor who says "we will not participate in this IPO" adds to the pressure the way GM's statement did in 2012. Post it. Put out a statement this week, not in next year's annual letter, when it will be too late. Tell your peers.

If you sit on a pension, endowment or foundation investment committee, ask your managers the same questions in writing, and ask how they will meet their fiduciary duty when an index provider changes its rules to make a purchase automatic.

And if you have relationships with elected officials, state treasurers or attorneys general, now is the time to pick up the phone. Tell them what's in the S-1. Most of them haven't read it.

We are not against AI. We are against selling an uninsurable risk to every retirement account in the country before anyone has shown it can be controlled. Once these shares are in the index funds, the risk is socialized for good. Before that, it is still a choice.

Whose money will carry it? Make sure the answer isn't yours.

Join us.

A sample email to your advisor

Copy it, change it, make it yours.

Subject: Please keep me out of the Anthropic IPO and Meta's Class C shares

Hi [Name],

I'm writing about the Anthropic IPO, which is expected to price in mid-November, and about the AI listings likely to follow it.

First, I don't want to participate in this IPO in any way: no allocation, no first-day purchases, and no new funds bought on my behalf to capture it. Please pass my concerns along to Anthropic and to the underwriters. The company's own S-1 says its safety controls "have failed, been misconfigured, or been inadvertently disabled," describes a state-sponsored cyberattack carried out with its tools, and warns that for a catastrophic incident "no single insurer, and potentially the insurance industry in the aggregate," may have the capacity to cover it. The financials concern me too: hundreds of billions in locked-in computing commitments against revenue customers can cut at any time, and slowing growth. For the same reasons, please don't buy Meta's new nonvoting Class C shares for me.

Second, please tell me which of my current holdings could end up buying Anthropic or other AI IPOs, including index funds that follow the Nasdaq-100 or Russell indexes and their new fast-entry rules. I'd like to see options that avoid automatic buying until a new listing has a year of trading and audited results.

Third, I'm considering saying publicly that I won't participate. I'd like your help understanding how to do that with my accounts in order.

I'd appreciate a reply by [date], since the roadshow is expected to start November 9.

Thank you, [Your name]