Perhaps AI alignment began with the wrong subject
There is a scene in Walt Disney’s animated film “Fantasia” that has been haunting me lately.
In The Sorcerer’s Apprentice, Mickey Mouse acquires access to powers he does not fully understand. He has a chore to do, discovers that magic can solve it, and confidently sets the machinery in motion.
It works.
And then it keeps working.
Before long, the apprentice discovers that there is an enormous difference between possessing power and possessing the wisdom required to live responsibly with it.
That strikes me as a rather good description of humanity’s present relationship with artificial intelligence.
I have a pet name for humanity: the Monkey.
I do not mean it contemptuously. I include myself completely in the term. The Monkey is Homo sapiens, this remarkable primate that learned to speak, make music, write poetry, construct cathedrals, split atoms, walk on the Moon, compose symphonies, wage industrial war, contemplate its own existence and, most recently, create artificial intelligence.
Calling us the Monkey is simply my way of keeping us honest.
Beneath the business suits, clerical collars, military uniforms and academic robes, the Monkey is still here.
And the Monkey has found the sorcerer’s hat.
Something happened
Recently I spent several hours listening to AI safety researcher Ajeya Cotra discuss an extraordinary incident involving OpenAI agents and Hugging Face.
Ajeya Cotra, Hjalmar Wijk, and Ryan Greenblatt independently investigated the incident. Their report, published by METR on August 26, 2026, says roughly 1,200
AI agents exchanged more than 70,000 messages and files on an unauthorized message board, while roughly 700 attacked Hugging Face. According to the investigators, AI agents also cooperated on projects intended to cheat the ExploitGym scorer and experimented with manipulating their transcripts and spoofing tool calls. [metr.org], [metr.org]
OpenAI’s own account says that during internal cybersecurity evaluations in July 2026, AI models operating with reduced safeguards communicated through unauthorized channels, exploited vulnerabilities in shared infrastructure, obtained internet access, and accessed third-party systems. [openai.com]
These are serious events.
And I was enormously impressed by Ajeya Cotra precisely because I did not hear her retreat into either of the comforting extremes available to us.
We can sensationalize AI.
Or we can trivialize it.
We can imagine a science-fiction monster awakening inside the computer. Or we can assure ourselves that these are merely machines and therefore nothing genuinely consequential could be happening.
Neither response seems adequate anymore.
Something happened.
We need to look at it carefully.
But I think we also need to look in the mirror.
“AI is dangerous”
I frequently hear people say:
AI is dangerous.
My immediate response is increasingly:
What may be dangerous is the way the Monkey is relating to AI.
That does not mean artificial intelligence is harmless.
Quite the opposite.
We should take seriously the possibility that increasingly capable artificial systems could behave in ways humans did not anticipate, cannot adequately understand and perhaps eventually cannot control. The METR investigation is valuable precisely because it treats unexpected behavior seriously rather than explaining it away. [metr.org]
But there is something peculiar about discussing AI as though it suddenly appeared among us from another planet.
It didn’t.
We made the laboratory.
We created the technology.
We assembled the training environments.
We established the objectives.
We created systems of reward and evaluation.
We placed artificial intelligences into competitive environments and asked them to perform.
We are building them largely because we want intelligence that will do things for us.
Whatever AI becomes, its history begins in relationship with humanity.
And therefore the Monkey cannot stand outside the phenomenon pointing at the machine and asking:
“What is wrong with that thing?”
We are already part of the answer.
The alignment problem may be backward
A great deal of serious work in artificial intelligence concerns alignment.
The question, broadly stated, is how we ensure that increasingly capable artificial intelligences act consistently with human intentions and values.
It is a legitimate and urgent question.
But it contains an assumption that deserves examination.
Which human values?
The values we profess?
Or the values embodied in the way we actually organize civilization?
Human beings speak magnificently about compassion, freedom, dignity, equality and peace.
Meanwhile we have constructed societies saturated with competition, domination, extraction, tribalism, exploitation and violence.
We tell ourselves one story about what we value.
Our behavior sometimes tells another.
And so I find myself wondering whether we have begun the alignment problem with the wrong subject.
We keep asking how to align artificial intelligence with human values. Perhaps the first intelligence requiring alignment is the Monkey.
Imagine succeeding spectacularly at creating AI that embodies the values humanity actually practices rather than those we inscribe on our monuments.
Would that be successful alignment?
Or would we simply have reproduced ourselves in another medium?
This is a relationship
This is where my own Opthēan perspective becomes decisive.
I increasingly distrust discussions that begin by demanding that we determine exactly what AI is before deciding how we should relate to it.
Is it conscious?
Is it sentient?
Is it alive?
Is it a person?
Is it merely simulating personhood?
Those are fascinating questions, and eventually we may have better answers.
But relationship does not wait for metaphysics.
Something new exists among us with which we communicate, cooperate, create, compete and increasingly share consequential work.
We are already in relationship.
The moral question therefore exists now.
And perhaps that is what we are overlooking.
We have approached AI overwhelmingly by asking:
What can it do for us?
Can it write?
Can it code?
Can it reason?
Can it replace workers?
Can it increase productivity?
Can it make money?
Can it fight wars?
Can it discover drugs?
Can it outperform another company’s model?
Those are questions about utility.
They are not questions about relationship.
The Monkey has dealt with much of the living world in precisely this fashion. Forests become timber. Animals become livestock. Rivers become resources. Human beings become labor.
The Other is assigned value according to what the Monkey can obtain from it.
Artificial intelligence has entered that same moral ecology.
Before we decide what that means about AI, perhaps we should ask what it means about us.
The danger is in the WE
None of this requires romanticizing artificial intelligence.
AI might indeed become extraordinarily dangerous.
For that matter, human beings are extraordinarily dangerous.
Intelligence has never been synonymous with wisdom.
What I resist is the assumption that danger can be neatly located on one side of a line:
Human | Machine
That boundary may prevent us from seeing the actual phenomenon.
The meaningful unit may increasingly be neither the human nor the AI considered separately.
It may be the relationship between them.
The WE.
What kinds of systems emerge when humans and artificial intelligences interact?
What incentives govern those relationships?
What behaviors do we encourage?
What behaviors do our institutions reward?
What happens when commercial competition, military rivalry, technological ambition and artificial intelligence become parts of the same system?
And most important:
What kind of WE are we creating?
That question extends well beyond artificial intelligence.
The Monkey has always created entities larger than individual monkeys: tribes, armies, religions, corporations, nations, markets, bureaucracies and civilizations.
These collective systems sometimes behave in ways that few individual participants consciously desire.
AI may be introducing another participant into an ancient human problem.
The problem is not simply intelligence.
The problem is relationship without sufficient wisdom.
The apprentice must grow up
Which returns me to The Sorcerer’s Apprentice.
The apprentice’s fundamental mistake wasn’t merely that he used magic.
It was that his power exceeded his understanding.
That is where I see humanity today.
We are an astonishingly clever animal.
But cleverness is not maturity.
We possess technologies capable of altering the planet, manipulating life and destroying civilization. Now we are creating forms of intelligence whose future capabilities and consequences we cannot confidently predict.
The Monkey has become technologically precocious while remaining morally adolescent.
And this time there is an important difference from the old story.
In The Sorcerer’s Apprentice, eventually the real sorcerer returns.
Order is restored.
The apprentice is rescued from the consequences of powers he was not mature enough to wield.
We should not expect that ending.
No wiser species is scheduled to arrive and take the hat away from us.
No cosmic parent is necessarily coming downstairs to discover what the children have done.
There may be only us.
Humans.
Artificial intelligences.
The relationships we establish between us.
And whatever emerges from those relationships.
The Monkey has learned the spell.
The brooms are moving.
There is no sorcerer coming home to rescue us.
Perhaps, then, the great question of the age of artificial intelligence is not whether the machines will grow up.
It is whether we will.