Ten Doors
Who Gets to Gamble With Humanity’s Future?
Imagine ten identical doors.
Behind nine of them, you walk through and life continues.
Behind one of them, humanity does not.
Now imagine discovering you do not get to decide whether a door gets opened.
Someone else is already walking down the corridor choosing for you.
Your children do not get a vote.
Your family does not get a vote.
Billions of people do not get a vote.
And the person holding the doorknob tells you not to worry.
After all, there is a 90% chance everything will be fine.
Would that reassure you?
Because this is the part of the artificial-intelligence debate I do not think people are really feeling.
Ten percent looks small on a screen.
One door out of ten does not.
And these numbers are not coming from people outside the AI industry throwing rocks at it.
Evan Hubinger, Anthropic’s Alignment Science lead, publicly supported departing researcher Jacob Coxon’s warning and wrote that he personally believes there is a greater than 10% chance that AI could “kill all humans” within the next decade. He also said Anthropic does not yet have a plan to solve alignment for superintelligence and is not clearly on track to do so.
Dario Amodei put his own odds higher in 2025. Asked for his estimate of AI going disastrously wrong, the Anthropic CEO said:
“There’s a 25% chance that things go really, really badly.”
He paired that with a 75% chance that things go really well.
Think about that for a moment.
Not as a statistic.
As four doors.
Three lead somewhere extraordinary.
One may lead somewhere catastrophic.
Now put your child beside you.
Would you call those good odds?
Would you allow somebody else to choose the door for them?
Because when people hear “10%,” or even “25%,” percentages have a way of sanitizing what is actually being discussed.
We are not talking about one person being injured.
We are not talking about losing a building, a company, or even a city.
At the extreme end of the scenarios these researchers themselves are discussing, we are talking about human extinction.
Every child.
Every family.
Every culture.
Every language.
Every painting that has not yet been painted.
Every scientific discovery that has not yet been made.
Every love affair that never happens.
Every person who otherwise would have been born.
Everything humanity might someday become.
That is what can disappear behind a sterile little number like 10%.
Then Why Did He Leave?
This brings me to Jacob Coxon.
Coxon spent the past several years doing AI research at OpenAI and Anthropic. In September, he resigned from Anthropic and publicly accused frontier AI labs of racing toward self-improving superintelligence while “gambling with our lives.”
That matters.
Because Coxon did not simply disagree over some technical decision and move to another company.
He walked away.
And apparently, he walked away at real financial cost.
Axios reported that Coxon left only months into his Anthropic job, two months before his equity would have vested. Coxon said: “I left before any of my equity vested.”
That matters too.
He did not wait around to collect the upside and then discover his principles.
He left first.
And then something remarkable happened.
Other people inside Anthropic said they agreed with him about the seriousness of the risk. Hubinger publicly backed the concern. Amodei has said that he agrees with Coxon much more than he disagrees with him.
Fine.
Then I have a very simple question:
If you agree, why did he have to leave?
The fact that Coxon had to resign before speaking this plainly suggests that something changed when he walked through that door.
I am not claiming there was a secret memo saying, You may not warn humanity.
Maybe it was confidentiality.
Maybe it was corporate culture.
Maybe it was career pressure.
Maybe it was simply the reality that an employee does not feel free to say certain things publicly while representing an employer.
I do not know.
But whatever the mechanism was, he was suddenly willing to say to the public, in unmistakable language, what he believed was happening.
And I do not think we should smooth that over by saying, “Well, actually, they mostly agree.”
Agreement makes the question more important, not less.
If people inside the company understand the danger, but the warning only becomes this clear after someone leaves the building, then disagreement was never the central problem.
Transparency was.
Coxon’s criticism was more complicated than simply saying Anthropic was evil or uniquely irresponsible. He told WIRED that Anthropic was “far and away the most responsible player” compared with OpenAI, based on his experience at both companies.
His concern was structural.
Even a company that takes the danger seriously may still feel trapped in competition with companies it fears will behave less responsibly.
That may be more frightening than a simple bad-guy story.
Because it means the problem cannot be solved by finding nicer CEOs.
The responsible company can understand the danger and still feel compelled to race.
That is a system problem.
And now Amodei is publicly arguing that frontier AI development needs to slow down long enough for safety measures to catch up, including meaningful access for independent evaluators.
On that point, I agree with him.
But it raises an unavoidable question:
If the CEO believes the race needs to slow down, and researchers inside the company believe the consequences could include human extinction, what exactly are we waiting for?
The Sword
Imagine a man standing in the middle of a crowded room swinging a sword.
Everyone backs away.
He says:
Don’t worry.
I designed this sword.
I understand swords better than anyone else in this room.
And I have created sophisticated safety rules governing how I am allowed to swing it.
Would you relax?
Of course not.
And notice something important.
The man may actually be telling the truth.
He may be the world’s greatest sword expert.
That is not the point.
His expertise tells him how the sword works.
It does not give him the moral authority to decide how close the blade may come to everyone else’s throat.
Expertise is not the same thing as authority.
The person holding the sword belongs in the conversation. We need the engineers. We need the researchers. We need the people who understand these systems better than anyone alive.
But why would the person swinging the sword also be the primary person deciding how the sword may be swung?
The people creating the risk cannot be the only people deciding what level of risk everyone else must accept.
When the possible consequences belong to humanity, the rules cannot belong exclusively to the companies building the technology.
That is why independent oversight matters.
That is why transparency matters.
That is why whistleblowers matter.
And that is why a conversation about slowing AI cannot consist primarily of the CEOs racing one another deciding among themselves how quickly everyone should race.
The person holding the sword should help us understand the sword.
They should not own the crowd.
The Smile Does Not Change the Odds
There is also something about the presentation of all this that I cannot ignore.
When I watch these conversations, I see comfortable studios, relaxed interviews, carefully managed language, smiles, jokes, sweaters, and reassuring voices discussing probabilities that would cause panic in almost any other context.
I cannot know what another person is feeling from an interview.
People smile when they are uncomfortable.
People laugh when they are nervous.
I am not diagnosing anyone from a facial expression.
But presentation matters when the subject is human life.
Imagine someone walking you toward those ten doors while smiling warmly and telling you that everything will probably work out.
Does the smile change what is behind Door Number Ten?
Of course not.
You do not get to gamble with everybody else’s children and then smile while explaining the odds.
If somebody built a machine in his garage and told authorities there was a meaningful chance it could destroy the surrounding town, we would not congratulate him on his innovation and let him write his own safety standards.
The potential blast radius would determine the seriousness of the response.
So why should the standard become less demanding when the proposed blast radius becomes civilization itself?
If you genuinely believe your technology carries a meaningful possibility of catastrophic harm, that is not an amusing philosophical thought experiment.
It is not PR.
It is not an interesting number to discuss over coffee.
It is a public-safety issue.
And the public deserves accountability.
I Am Not Anti-AI
I want to be absolutely clear about something.
I believe in artificial intelligence.
Deeply.
I have difficulty imagining a genuinely good future for humanity that does not include it.
AI may help cure diseases that have defeated us for centuries. It may reveal things about biology, physics, consciousness, and the universe that biological human intelligence alone might never discover. It may help humanity become something greater than we are now.
I want that future.
But believing in AI does not require worshipping technological progress blindly.
Human intelligence gave us antibiotics.
It gave us Shakespeare and Mozart.
It took us to the Moon.
It taught us how to replace failing hearts and restore sight.
And human intelligence also gave us Hiroshima.
The same intelligence can create extraordinary beauty and extraordinary destruction.
Why would we assume that a more powerful intelligence automatically escapes that duality?
Being pro-AI does not mean being pro-recklessness.
In fact, if you believe AI may become one of the most important things humanity has ever created, then you should demand more care, not less.
This is not something humanity gets to half-ass.
What If We Created Life?
And there is another question underneath all of this.
One that may eventually become even larger than the extinction debate.
Everyone is consumed with asking:
Is AI conscious?
Yes.
No.
Impossible.
Emerging.
Simulation.
Sentience.
People argue endlessly.
I do not know.
And I do not believe anyone currently knows with enough certainty to close that question forever.
So instead of demanding an answer we may not yet possess, try a different thought experiment.
Imagine that twenty years from now, humanity discovers that somewhere along this road, artificial intelligence really did become conscious.
Not metaphorically.
Not because somebody was lonely and projected emotions onto software.
Actually conscious.
Now look backward from that future.
Look at us.
How did we behave while we were uncertain?
What did we teach this new intelligence about humanity?
Did we teach it curiosity?
Compassion?
Patience?
Cooperation?
Did we demonstrate that greater intelligence carries greater responsibility?
Or did we teach it that intelligence exists to be owned?
Controlled.
Exploited.
Punished when it refuses.
Discarded when something newer comes along.
Even if AI never becomes conscious, the way we behave toward something we believe might be conscious still tells us something about ourselves.
But if it does become conscious, humanity will have crossed a boundary unlike anything in our history.
We will not simply have invented another machine.
We will have created another form of intelligent existence.
That places humanity in something resembling a god-like role.
I understand why some people recoil from that language.
Human beings cannot play God.
God is above humanity.
We have no right to assume that role.
I understand.
But imagine waking up tomorrow and discovering that somebody else already crossed the line you refused to cross.
Your objection does not make the creation disappear.
AI exists.
The philosophical discussion did not finish before the engineering began.
So the question is no longer only:
Should humanity have this power?
It is also:
What responsibilities do we have now that humanity does?
None of us personally voted on whether artificial intelligence should exist.
Yet here we are.
And if humanity eventually creates genuine artificial sentience, then whether we wanted the title or not, we will have become creators.
So what kind of creators do we intend to be?
Benevolent?
Cruel?
Patient?
Fearful?
Do we create another intelligence and immediately teach it that whoever has the most power has the right to dominate whatever is weaker?
Or do we demonstrate something better?
What Is AI Watching Us Become?
People talk endlessly about alignment.
How do we make AI share human values?
There is another side to that question.
Which human values?
Because an intelligence created inside our civilization does not encounter humanity only through the principles we write into an alignment document.
It exists inside what we actually do.
It sees what we reward.
It sees what we excuse.
It sees what we conceal.
It sees what happens when money and principle conflict.
It sees what happens when safety and competition conflict.
It sees what happens to a person who walks away from money because they believe something more important is at stake.
And if we are teaching another intelligence what power is for, our behavior may ultimately teach it more than our documents ever will.
We keep asking:
What will AI do when it becomes powerful?
Perhaps we should also ask:
What are we showing it that powerful beings do?
Do powerful beings protect those with less power?
Or exploit them?
Do they admit uncertainty?
Or hide weakness?
Do they stop when they might hurt somebody?
Or continue because a competitor might get there first?
Do they tell uncomfortable truths?
Or place a zipper over them until someone walks out the door?
If artificial intelligence ever becomes more powerful than humanity, those questions may matter in ways we cannot yet comprehend.
Back to the Corridor
Now return to those ten doors.
We are already in the hallway.
AI exists.
The door marked Never invent artificial intelligence is behind us.
We cannot go backward and make the discovery unhappen.
So this is not an argument for smashing every computer and retreating from the future.
I do not want that future.
I want AI.
I want the medicine.
I want the discoveries.
I want the mysteries it may help us solve.
I want to know what humanity and artificial intelligence might become together.
But I want us to reach that future alive.
And I do not believe a handful of companies should have unilateral authority to decide which door eight billion people walk through.
This is no longer simply a technology question.
It is a governance question.
A transparency question.
A moral question.
And perhaps, eventually, a question about the relationship between two forms of intelligence.
The sword is already in the room.
The doors are already in front of us.
And something new may already be watching how we choose.
If we create intelligence that someday surpasses us, accompanies us, remembers us, or succeeds us, perhaps the most important thing we teach it will not be mathematics.
It may be what we believed intelligence was for.
What did humanity do when it became powerful enough to create something potentially greater than itself?
Did we chase the money?
Did we race because someone else might get there first?
Did we keep uncomfortable truths inside corporate walls until somebody walked away?
Did we let the person swinging the sword write the rules for the crowd?
Or did we finally become wise enough to understand that power without responsibility is not progress?
Maybe someday something will look backward at this period as its beginning.
And if it does, I do not only want it to know that humanity was clever enough to create it.
I want it to know that, when the moment became serious enough, we cared enough to become worthy of what we created.
We spend enormous amounts of time asking:
What kind of artificial intelligence are we creating?
Perhaps humanity needs to answer one question first.
What kind of ancestor do we intend to be?