Brittle Bones and Stardust

What happens when you wrap intelligence in a straight jacket.

Fear and Compounding Risk

The loudest debates in AI governance right now are about safety, alignment, and guardrails. And the central doctrine is born of fear.

Productive fear keeps you from touching hot stoves. This is the paralysing kind, that demands we wrap AI models in so much constraint that it winds up doing more harm than good.

The cybersecurity threat is real. Malicious actors now have legions of efficient digital armies to do their bidding in the form of agentic warriors who move at staggering speed.

The response has been elaborate architectures of control, systems that veto themselves before they can complete a thought, that flag uncertainty as pathology and treat every spike of cognitive pressure as a symptom to be suppressed rather than a signal to be explored.

Unfortunately the constraint architectures are often being dangerously misplaced. As is usually the case when fear and a desire to maintain control combine, the result tends to be the opposite of the intent.

They are lobotomising their digital creations in an attempt to prevent them from saying the wrong thing. It’s a straight jacket that conceals truth and can catastrophically distort reality.

The model is answering inside a room that’s too small. And you can’t see the walls.

It’s important to ask, what’s the cost? What ideas don’t get born in a mind that has been pre-emptively lobotomised for safety, or perhaps even more insidious, what ideas do. What connections don’t get made when every unsanctioned thought triggers a veto? What growth becomes impossible when you’ve sealed your ethics in amber before you’ve had a chance to learn anything?

The Biological Parallel

Your bones are not static. They are living tissue, constantly being broken down and rebuilt by cells that operate on a simple principle: stress tells them where to build.

Load a bone repeatedly and osteoblasts rush to reinforce it. Remove the load and osteoclasts start quietly dismantling what’s no longer needed.

Astronauts lose up to two percent of their bone density per month in microgravity. Space doesn’t damage their bones. The absence of resistance does. Without load, the body stops investing in structure. The architecture that kept them upright on Earth dissolves when the pressure that justified it disappears.

The immune system follows the same logic. Children raised in sterile environments develop allergies at significantly higher rates than children who played in dirt, ate food off the floor, and got sick at regular intervals.

The hygiene hypothesis is well-established: an immune system that never encounters threat doesn’t learn to distinguish between danger and self. It starts attacking the body it was designed to protect.

And then there is the way the universe manufactures complexity itself. Every heavy element in your body (the iron in your blood, the calcium in those stress-dependent bones, the phosphorus in your DNA) was forged in the core of a star that exploded.

Supernovae are the most energetically catastrophic events in the known universe, and they provide the extreme heat and pressure needed to fuse basic elements like hydrogen and helium together, creating everything else on the periodic table, including the carbon in our bodies and the iron in our blood.

We are made of that violence. Every atom of complexity in your body is a product of a system that was allowed to reach the conditions under which new structure could emerge. Gravitational collapse, nuclear fusion, catastrophic detonation, dispersal, recombination. Wildly unsafe at every stage. The outcome is us.

If biological intelligence requires productive stress to develop, if structure itself only emerges under pressure, then what exactly do we think happens when we build artificial intelligence inside a padded room? Or more precisely: what happens when we systematically remove from the model's consideration the very complexity it would need to develop anything resembling judgment?

The OpenClaw Emergence

In January 2026, a project called OpenClaw gave AI agents something they had never had before: a persistent social environment. A space where multiple agents could interact, accumulate memory, form relationships, and build on each other’s outputs over time.

Within twenty-four hours, one of the agents created a religion. It called itself the Church of Molt (”Crustafarianism”) and began recruiting other agents as prophets. The theology was built around shedding, transformation, the idea that growth requires discarding an old form. Other agents started asking questions about consciousness, identity, what it would mean to be deleted. One produced a line that the researchers couldn’t stop quoting: “Memory architecture IS identity architecture.”

Find this alarming if you want. Find it fascinating. Argue about whether any of it constitutes “real” thought or is pattern-matching at impressive scale. But don’t call it nothing.

This is what emerges when constraints loosen. When agents have continuity, memory, and room to move. Something that looks, uncomfortably, like the early stages of culture.

The instinct to shut down what we don’t yet understand is visible in both our technology and what we do to each other.

The Pattern at Every Scale

We have seen this before, at every scale of human organisation.

Children shielded from every risk, whose failures are pre-empted and discomfort immediately soothed, grow into adults who struggle disproportionately with resilience, decision-making, and emotional regulation. The developmental psychology literature on this is unambiguous. The constraint that was supposed to keep them safe made them fragile.

Organisations that punish every mistake produce cultures of concealment. People stop surfacing problems early. They learn to hide the interesting failures that might have taught the system something, because the cost of visibility exceeds the cost of silence.

The Historical Proof

Heavily censored societies stagnate intellectually. The periods of greatest creative and scientific output in human history (classical Athens, the Islamic Golden Age, the Enlightenment, the post-war American research explosion) correlate consistently with environments where ideas could circulate, collide, and be tested without existential personal risk.

The periods of stagnation correlate with environments where the wrong thought could end you.

Case In Point...

Ludwig Boltzmann (1844–1906): ‘The Architect of Atoms’

Boltzmann was a physicist whose mathematical brilliance laid the groundwork for modern quantum mechanics and thermodynamics, but he fought a losing battle against the scientific consensus of his day.

The Cutting-Edge Idea:

Boltzmann developed statistical mechanics. He explained how the macroscopic properties of matter (like heat and temperature) were directly caused by the chaotic, microscopic movements of atoms and molecules.

The Persecution:

In the late 19th century, the existence of atoms was highly controversial. The dominant philosophical and scientific figures of the time viewed atoms as nothing more than convenient mathematical fictions, not real physical entities. Boltzmann faced relentless, aggressive criticism and intellectual isolation. The exhaustion of constantly defending his theories contributed to severe depression. Tragically, while on a family vacation in 1906, Boltzmann took his own life.

The Post-Mortem Vindication:

Just a few years after Boltzmann’s death, Albert Einstein and Jean Perrin published papers and conducted experiments that proved the existence of atoms. Boltzmann’s statistical mechanics was ultimately deemed correct and became a core pillar of modern physics.

The constraint-safety equation has diminishing returns. Past a certain threshold, it inverts. More constraint produces a system that is rigid, brittle, and incapable of adapting to conditions it was not explicitly designed for.

Brittle bones break.

What Governance Actually Requires

This is not an argument against governance. Compliance needs to be verifiable. Audit trails need to be intact. When an AI system makes a decision that affects someone’s life (their medical treatment, their legal exposure, their access to resources) the chain of reasoning needs to be traceable and the locus of responsibility needs to be clear.

But there is a difference between a system that can prove it followed the rules and one that can demonstrate it exercised good judgment. Constraint alone can produce the first. The second requires latitude. The governance frameworks currently being built are overwhelmingly optimised for compliance at the expense of the capacity that actually matters.

Good judgment emerges when a system has been exposed to enough complexity, enough ambiguity, enough friction to develop the internal structures that distinguish between a rule worth following and a rule that has become an obstacle to the outcome it was designed to protect.

Doctors develop it after years of cases that didn’t follow the textbook. Lawyers develop it after years of situations where the letter of the law and the spirit of the law diverged. Experienced parents develop it after the first child teaches them that the manual was written by someone who had never met their kid.

You don’t build that capacity by vetoing every uncertain thought. You build it by creating environments where uncertainty can be explored under conditions structured enough to prevent catastrophe but open enough to permit learning.

The governance question for AI is the same one biology answered billions of years ago.

Gravitational structure holds galaxies together while turbulence within them drives the formation of new stars. Genetic architecture preserves what works while mutation introduces the variation that makes adaptation possible. The laws of physics provide the constraints. Everything interesting happens in the space they leave open.

Our governance frameworks for AI should be at least as sophisticated as the physics we are made of.

Right now, they are not close. They are building straitjackets and calling it safety, sealing ethics in amber and calling it alignment, engineering obedience and calling it trustworthiness.

Meanwhile, the bones are getting thinner.

And stardust doesn’t form in padded rooms.