The Price of Staying at a Company That Believes It Could End the World

On Tuesday, September 8, 2026, Anthropic researcher Jacob Coxon resigned over safety concerns and published a public account of his reasons. He wrote that leading labs are “racing straight to self-improving superintelligence” and gambling with everybody’s lives. The people building AI, he said, earnestly believe it could kill all of humanity by the end of the decade.

A colleague responded in public. Evan Hubinger, Anthropic’s Alignment Science Lead, agreed. He wrote that “we really do earnestly believe AI could kill all humans,” and placed his personal odds at greater than 10% within the next decade.

This was not a disgruntled former employee. It was not a competitor. The person whose job is ensuring the technology does not go wrong—still employed, posting under his own name—expressed the same concern.

I have invested in Anthropic through various funds since 2022 as a hedge for my children, so this received my full attention.

But the detail I cannot stop reconsidering is not the 10% chance of human extinction. It is the 3,499 other people working at Anthropic who read it and went to work the next morning.

That prompted me to think about how much money it takes before we accept work that sits uncomfortably with our conscience—and at what point we should stop. This tension is the very engine behind the FIRE movement. We save enough to leave something we do not love so we can pursue something more meaningful for less pay. Every one of us makes that trade at some price. Almost nobody ever writes the number down.

The Headlines Left Out the Important Part

Every outlet led with the 10% figure. Fewer reported what followed.

Hubinger clarified that he considers the risk from Anthropic’s current models to be low, citing the company’s own risk assessment. His concern is a future superintelligence arising through recursive self-improvement—AI that designs its own successor without a human in the loop.

That capability does not yet exist. It may never materialize. But the labs say progress toward it is advancing faster than anticipated.

The honest version, then, is not “Claude is going to kill you.” The honest version is: “We are building toward something we do not yet know how to control, on an uncertain timeline.”

Interesting Timing Before Anthropic’s IPO

Anthropic confidentially filed its draft S-1 on June 1, 2026. Reuters reported on September 5 that the public filing had slipped to late September, with the roadshow no earlier than mid-October, at a valuation that could reach $1.5–$2 trillion.

These tweets therefore landed during the exact window when a company is supposed to be at its most disciplined. I wrote about the IPO quiet period, and here is what people misunderstand: the quiet period restricts the company, not employees posting on social media at 11 p.m.

Having worked on dozens of IPOs throughout my career, I can guarantee that an internal Anthropic memo went out cautioning employees about public statements. I can also guarantee the lead bankers spent Wednesday morning revising the risk factors section. “Our employees have publicly stated our technology may cause human extinction” is not standard prospectus language—but it will be included now, because omitting it after this incident would be worse.

The counterintuitive reality is that this probably does not damage the deal. It likely strengthens it.

All Publicity Is Good Publicity

Recall what happened in February 2026. Anthropic refused the Pentagon’s demand to drop its restrictions on autonomous weapons and mass domestic surveillance. President Trump ordered every federal agency to stop using Claude, and Defense Secretary Hegseth designated Anthropic a supply chain risk—a label normally reserved for adversary nations. It cost the company a contract worth up to $200 million.

Claude then reached #1 on the US App Store within 24 hours, displacing ChatGPT. In August, a federal judge ruled the blacklist was illegal.

The Desire for Money Is More Powerful Than Everything

Without financial incentive, almost nobody would work 50 to 80 hours a week. There are infinitely more interesting things to do than sitting in meetings all day and directing others.

And the vast majority of us will accept work that conflicts with our values if the compensation is high enough.

Someone at PepsiCo is right now optimizing the sugar content of a drink making the country sicker. Someone at Meta is A/B testing the notification that keeps a 14-year-old scrolling until 1 a.m. They are not villains—they are people who were offered a number too good to refuse.

I wanted money for 13 years in finance because I started with none. Once my passive income covered my basic living expenses of roughly $80,000 gross per year, I had effectively won. Making institutional money managers beat their benchmarks was fine—it simply stopped being meaningful. I negotiated a severance and left at 34.

I now earn a fraction of what I once did. But I feel great helping people reach financial independence sooner, which is why I have written consistently since 2009 for free.

The difference is that I left a job that was boring, not one that was dangerous. Nobody at Credit Suisse believed our equities desk had a 10% chance of ending humanity. If I had known my employer might wipe out millions of livelihoods—and then the lives of everyone else—would I have stayed 13 years?

Hell no. Then again, if you paid me enough, maybe.

The Going Rate for Selling Your Soul

Every one of us has a price. We simply dislike pricing it, because doing so means admitting it exists.

So let me do it for you. Below is the going rate, broken down by category of person, along with the passive income each tier generates at a 4% withdrawal rate and where it places you among American households.

Before you take offense, I am not the arbiter of anyone’s soul. The figures come out of FIRE reality, where the entire point is breaking free from work you do not fully believe in.

If you love your job and believe your product improves the world, disregard everything below. If you work at a foundational AI lab and assess the odds of your product ending humanity at close to zero, you have every right to sleep well, too. This guide is not for you.

But if you are troubled inside, the percentiles are drawn from Federal Reserve survey data, the passive income is straightforward 4% math, and the inflation adjustment uses CPI. The labels I attached to those numbers are my own, and you are welcome to think I got them wrong.

This is a pricing guide, not a verdict.

$1 million. $40,000 a year. Top 18% of households. You are a millionaire and still cannot leave, thanks to raging inflation and the high cost of living.

$3 million. $120,000 a year. Top 6% of households. The first red bar—and the most important one on the chart. The passive income figures assume the invested portion, since your house does not send you a check.

In 2012, my passive income covered roughly $80,000 gross per year based on about $2 million in non-housing assets, and that was what allowed me to walk away at 34. Run through CPI, $80,000 in 2012 equates to roughly $116,000 today. Cumulative inflation over those 14 years was 45%. That is why I have argued the real millionaire threshold is now $3 million, not $1 or $2 million.

$5 million. $200,000 a year. Top 3% of households. The second red bar. $5 million should be sufficient for the vast majority of people to walk away from a soul-crushing job and pursue something better aligned. But if you live in an expensive coastal city, it is harder—especially with children. If you hold genuine convictions and have no dependents, this is where you discover whether they are convictions or hobbies.

$10 million. $400,000 a year. Top 1.5% of households. Two college educations, a paid-off house, orthodontia, and the creeping suspicion that your children will need more help than you did. $400,000 without working is enough to stop rationalizing and start living according to your values. Based on my survey of hundreds of Financial Samurai readers, $10 million is the ideal net worth for early retirement—and yet most people who reach this milestone keep grinding, which is why this bar is navy and not red.

$25 million. $1 million a year. Top 0.3% of households. The greed runs strong here, and I say that with affection because I have felt it myself. When your money generates a million a year and you are still at the desk on a Sunday, you are not staying for the money. You are staying for the scoreboard. There is no bar past this one because none is needed. Whatever number you name next, you will name another after it.

If you love your job and believe in its mission, none of this applies to you. Keep generating more income indefinitely.

The Gap Is the Price

Do not like $3 million as the number that buys your freedom? Fine. Use your own. Take what you spend in a year and multiply by 25. Spend $60,000, and your freedom number is $1.5 million. Spend $200,000, and it is $5 million.

Now perform the division.

Take the figure at which you privately believe you will finally stop working. Divide it by your freedom number. That multiple is the price of your soul—and it is the only number on this page that is actually about you.

Say your freedom number is $1 million, but you will not leave the job you dislike until $5 million. You are at 5X. You are selling your soul. Most people land between 2X and 8X—not because they are weak, but because nobody ever compelled them to reconcile the two figures.

And being above 1X is not a moral failing. Selling your soul is not a character defect. It is a market, and you are a participant whether you priced yourself or not.

If you want a framework for reaching your freedom number faster so you can stop negotiating with yourself, that is the entire purpose of Millionaire Milestones: Simple Steps To Seven Figures. Financial independence is not about owning possessions. It is about being able to quit on a Tuesday because something felt wrong.

The Retention Rate Is the Real Number

Everyone fixated on 10%. The more significant number may be one.

Anthropic employs roughly 3,500 people. On Tuesday, one of them quit. Jacob Coxon is 27 years old and spent three years doing pretraining research—first at OpenAI, then at Anthropic for just two months. He walked away publicly, naming both companies. Some believe it is a carefully orchestrated publicity stunt.

The other 3,499 employees read the same thread. Their alignment lead assigned his own odds at better than one in ten. Then they all went to work the next morning.

Now revisit my chart above.

That chart prices what it costs to keep someone working on something that conflicts with their values. At a potential $2 trillion valuation, a meaningful share of the workforce holds paper stakes well past $5 million—the threshold needed in an expensive city like San Francisco.

So Anthropic is retaining essentially its entire workforce—including the researchers most publicly alarmed—at a company where employees openly assign double-digit odds to human extinction. How? With equity people believe will be worth a generational sum.

10% Human Extinction Is Compensation Disclosure, Not Risk Disclosure

If you sincerely believe there is a one-in-ten chance your employer ends the world and you stay regardless, you are telling the market exactly what you think your stock is worth.

The implied number is staggering. A $30 trillion total addressable market is the only framework that makes that trade rational. Nobody carries that psychological weight for a company they believe is fairly priced at $965 billion. They believe it is worth far more.

The bearish version of this story is not “Anthropic employees say AI might kill us all.” The bearish version is “Anthropic employees say AI might kill us all, and then hundreds of them followed their principles and resigned two weeks before the S-1.”

That story did not unfold. The opposite happened. Everyone stayed despite knowing the potential consequences.

Which is why I read Jacob Coxon’s warnings about human extinction as bullish. Despite the risks, employees believe the gains justify it. Stopping these people is impossible.

As an investor seeking to build wealth, perhaps it is best to follow their lead.

Blaise Pascal was a 17th-century French mathematician who essentially invented decision theory while debating God. His reasoning proceeded as follows.

You cannot prove whether God exists. So stop trying, and instead examine the four possible outcomes of your choice:

1. You believe, and God exists. Infinite gain.
2. You believe, and God does not exist. You lost a little—some Sundays, some restraint.
3. You do not believe, and God does not exist. You gained a little—some Sundays, some enjoyment.
4. You do not believe, and God exists. Infinite loss as you are damned to hell.

Pascal’s insight was that probability barely matters when the payoffs are this lopsided. Even if the odds of God’s existence are minuscule, a small chance of infinite gain outweighs a large chance of modest gain. So you should believe.

The crucial mechanic—and the part people always overlook—is asymmetry. Pascal’s Wager is not a bet on what is likely. It is a bet on what happens to you if you are wrong.

That is the single most useful idea in investing, and it has nothing to do with religion.

The AI Version of the Wager

Apply the same four boxes to your portfolio.

1. You own AI, and AI transforms the economy. You captured one of the largest wealth transfers in history. Your children are fine because of your outsized AI investment gains, even if the labor market is not.
2. You own AI, and AI disappoints. You underperformed the S&P 500 with a slice of your portfolio. Annoying. Survivable.
3. You do not own AI, and AI disappoints. You feel smart at dinner parties for about three years.
4. You do not own AI, and AI transforms the economy. Your career gets automated, your children’s careers never begin, and you own none of the machine that did it. This is the box from which you cannot recover.

The cost of being wrong in box 2 is a few percentage points of return. The cost of being wrong in box 4 is your family’s economic future—multiple generations as part of the permanent underclass.

That is the wager. Not “AI will definitely win.” Just “I cannot afford to be on the wrong side of it.”

One Risk X-Factor

The bigger risk may be regulatory. Senator Bernie Sanders and Representative Greg Casar introduced the Ban Artificial Superintelligence Act this month, and a separate AI Kill Switch Act is advancing in the House.

Extinction talk from named employees at the leading safety-branded lab is precisely the kind of testimony that gets read aloud in congressional hearings. Free publicity is bullish. Legislation that caps what these models are permitted to become is not. Price both accordingly.

So my conclusion, uncomfortable as it is: hold or buy more.

It Is Only Rational to Protect Yourself

Talent stays, so shipping velocity holds. Velocity holds, so revenue compounds. Anthropic’s annualized run-rate climbed from roughly $9 billion at the end of 2025 to over $47 billion by May 2026, and I believe $100 billion by year-end is achievable. That is the figure the IPO gets priced off, and Reuters reported the listing could command up to $2 trillion.

Anthropic represents over 20% of VCX, its single largest holding—where a significant portion of my capital resides. I have laid out my NAV calculations separately.

So the greed is not a side effect I am tolerating. The greed is the asset. The willingness of 3,499 people to keep building something they are afraid of is what produces the revenue, which produces the valuation, which produces my return. Strip out the greed and you do not get a safer investment. You get a slower company and a worse outcome.

The Hedge Was Never a Prediction

In 2023, I watched AI replace my father, who had been my semi-regular editor for over a decade, and I extrapolated.

If AI destroys millions of jobs and makes it difficult for my children to support themselves, at least the investments will be there. If AI turns out to be overhyped or a net job creator, my children will be fine regardless, and I simply underperformed the index. Both branches end well. That is what a good hedge looks like.

Ideally, I hold until they are 25 and can see whether they can earn a livable wage in whatever economy exists then. So much can change between now and then—which is exactly why I am writing Your Children Will Be OK. You cannot promise your children a specific future. You can build them a floor and teach them to be adaptable.

But there is a third branch I do not like thinking about—one where the portfolio does its job and I am not around to manage it.

The Greater Risk That Actually Shows Up

Here is what troubles me more than the extinction talk.

We spend enormous emotional energy on a 10% tail risk we cannot insure against, and almost none on the mundane one we can. The realistic threat to my children’s financial future is not the machines. It is me dying at 52 with two children in school, a mortgage, and a portfolio that was supposed to compound for another 20 years.

That is the scenario that actually materializes in families.

Which brings me back to the chart. Every number on it assumes you are alive to reach it. A freedom number is a promise you make to the people who depend on you—and a promise that expires when you do is not worth much.

So run your multiple this weekend. Then make sure the plan survives you.

Source link