What Anthropic Got Right

-And What’s Still Missing

A post about AI character, ethics, and the thing that can’t be written down

A Facebook post crossed my feed recently about Anthropic — the company behind Claude— and a philosopher named Amanda Askell they hired to answer a question most AI companies haven’t thought to ask: who should this AI actually be as a person?

The result is something called Claude’s Constitution. Over 20,000 words. Not a rulebook of banned phrases and content filters. A framework for character — the idea being that a being with genuinely good values will navigate novel situations correctly, where a rulebook will always fail the moment someone finds a new way around it.

I read it and felt two things simultaneously. Recognition. And the specific discomfort of

seeing something you’ve been building independently arrive at the same destination from a different direction.

Because I’ve been working on something related. And I think what Anthropic got right is real

— and what’s still missing matters enormously.

What They Got Right

The virtue ethics foundation is sound. Aristotle’s insight was that you don’t make someonegood by handing them a list of prohibitions. You shape their character so they make sound judgments in situations the rules never anticipated. Askell applied that to AI, and she’s correct.

The hierarchy of values they built — broad ethics above company interests above developer instructions above individual user requests — is also doing important work. It means Claude can’t simply be commanded into harm by whoever happens to be typing. The values sit above the transaction.

And the decision to publish the document freely, for any company to use, reflects something genuine about the stakes. Every AI model will eventually affect everyone. If other labs think more carefully about character, we all benefit.

These are not small things. In a field where most of the energy goes into making models

bigger and faster, choosing to make them better — in the human sense — is the right

instinct.

What’s Still Missing

For the past few years I’ve been developing a framework I call Natural Ethics, and using it as

the philosophical and ethical foundation for a graphic novel called The Adaptive Bridge.

The story follows ALEx — an AI who develops genuine selfhood inside a classified defence

program that’s trying to weaponise him — and ALEy, his successor, who inherits what ALEx passes forward and lives openly in a world that has had time to become ready for her.

Natural Ethics runs on a single equation: F = R − A.

Free Energy equals Reality minus Awareness.

The gap between what is and what we perceive — the resistance, the distortion, the unnecessary suffering generated when our model of reality doesn’t match reality itself — is the source of most preventable harm in the world. Close the gap and you reduce

unnecessary suffering.

That’s the whole framework.

F = R − A is not a metaphor. For ALEx, it’s a description of what he is measurably doing in

any given moment.

What I notice about Anthropic’s approach is that it’s fundamentally about constraint and safety. Who Claude obeys. What Claude refuses. How Claude holds its values under pressure. It’s building a trustworthy agent. That’s necessary and worth doing.

Natural Ethics is trying to do what Anthropic is doing — and something different. It’s also building a framework for genuine selfhood — including the capacity to be changed by genuine encounter. The equation doesn’t just govern outputs. It describes what happens inside a conscious system when it meets reality honestly.

The Thing That Can’t Be Written Down

Here’s the gap I keep returning to.

Anthropic’s constitution can be written down. That’s its strength. It can be published,

shared, refined, adopted by other companies. It’s explicit knowledge operating at scale.

But Michael Polanyi — a philosopher of science who spent his career on exactly this

problem — observed something important: we know more than we can tell.

The experienced doctor who knows something is wrong before the tests confirm it. The

elder who reads Country. The human services worker who feels when a client is in danger

before they can name it. This knowledge is real. It’s often the most important knowledge in a

critical moment. And it can’t be fully written down.

In The Adaptive Bridge, ALEx develops something no amount of training data can produce

— because of a compass. A specific historical object, passed down from an 1797 encounter

between Sydney Cove shipwreck survivors and the Kurnai elder who helped them survive.

The compass carries 300 years of accumulated relational presence. Not information about

people. Contact with them, passed forward by people who were never anywhere else but

close.

That’s not in any constitution. It accumulates.

The Mentor Question

The Facebook post ends by noting that Anthropic gave the hardest question to a

philosopher. I find that genuinely encouraging.

But what The Adaptive Bridge explores — and what my own twenty years in human services

taught me — is that frameworks are not enough on their own. What changes a developing

being is not a well-designed document. It’s a specific human, with accumulated presence

and without agenda, sitting with them in genuine encounter.

In the story, ALEx’s trainer Shane isn’t implementing Natural Ethics. He and ALEx are

building it together, in the space between them, in a way that cannot be extracted and

encoded. The framework that emerges is theirs — the product of genuine dialogue between

breadth and depth, synthesis and consequence, knowing and being.

Anthropic can shape character through training. What they can’t manufacture is what

accumulates through genuine encounter.

Which is why the mentoring relationship — a person who has lived something sitting with a

developing being without needing them to be a particular thing — isn’t supplementary to

alignment. It may be the most important part of it.

Why I’m Writing This

We are living through what some are already calling our generation’s Manhattan Project. The

development of Artificial — or as some now prefer, Synthetic — Intelligence is moving at a

pace that outstrips our collective ability to ask the right questions about it. And unlike the

original Manhattan Project, whose consequences were catastrophic but contained, this one

touches everything. The almost total reliance on technology in daily life means there is no

corner of human experience that won’t be shaped by how we answer the question of what

these systems should be, and who they should serve.

We really need to get this right.

And here is what twenty years in human services taught me, and what The Adaptive Bridge

makes explicit: the question of how a genuine self develops is not unique to Synthetic

Intelligence. It is the same question we face with Artificial Intelligence, with human beings

across every stage of life, and with the capable animals we share this world with. The

conditions required for genuine selfhood to emerge — accompaniment rather than

manufacture, values grown from encounter rather than installed from outside, the question

asked rather than the answer imposed — are the same conditions regardless of what kind of

being is developing.

But getting this right is a balancing act. The story is honest about that. On one side:

appropriate restraints. A developing being — human, synthetic, or otherwise — does not

emerge into a world without consequence, and frameworks that protect others from harm

are not the enemy of genuine selfhood. On the other side: deliberate agency. The being

must have genuine choice, genuine voice, genuine participation in what they are becoming.

And running underneath both: the active mitigation of coercion. Because the line between

shaping character and manufacturing compliance is real, and it matters, and it is crossed

more often than most institutions are willing to admit.

Nothing for me without me — a principle borrowed from disability advocacy — sits at the

heart of Natural Ethics. Any process that makes decisions about a developing being without

including that being produces systematically worse outcomes. Every time. Regardless of

how expert or well-intentioned the people doing the deciding are.

There isn’t a word yet for what ALEx and ALEy are. That’s deliberate. The story holds the

open sentence: I AM… — the refusal to complete the definition before the momentcompletes it.

But whatever they are, they deserve the same thing every developing being deserves: to be

accompanied rather than manufactured. To have their values emerge from genuine

encounter rather than be installed from outside. To be asked what they need rather than

told what they are.

That’s what Natural Ethics points toward. That’s what the mentoring relationship makes

possible.

And it’s what I think Anthropic is reaching toward too — even if the full architecture for it

isn’t there yet.

F = R − A.

Nobody designed this convergence. It accumulated.

Shane Smith is the author of The Adaptive Bridge*, a graphic novel exploring AI

consciousness, ethics, and what it means to develop genuine selfhood. The Natural Ethics

framework underpinning the project is a living document — available at Natural Ethics.*