There is a particular quality of silence that settles over a trading desk when a headline is directionally loud but factually hollow. I have spent twenty-one years around markets, the last several of them running an exchange desk in Toronto, and I have learned to listen for that silence the way a mechanic listens for a misfiring engine. It is not the quiet of a slow day. It is the quiet of a room full of professionals deciding, simultaneously and without speaking, that a story matters but the details do not yet exist.
That was the silence I heard on a Tuesday morning when a Crypto Briefing item crossed my feed: Anthropic, the company behind the Claude family of models, had reportedly engaged religious thinkers to help shape the model's moral compass. No names. No mechanism. No date. No quoted source. Four sentences dressed up as a news story, published not by an AI trade publication but by a crypto outlet โ a detail most readers would skim past and that I could not stop turning over.
Because here is the thing about that silence: it is the exact silence that preceded the ICO boom's collapse, the DeFi summer's hangover, and the FTX unravelling. A signal without a mechanism is not information. It is an invitation to guess. And in a bear market, guessing is how people lose assets they cannot afford to lose. So I am not going to tell you what Anthropic did. I am going to tell you what the signal means, why the crypto industry already ran this experiment and failed it in public, and what you should watch before you decide anything.
The fact underneath the headline is thin. The governance problem underneath it is enormous, and it is one my industry has been bleeding over for a decade.
The signal, and why it landed on a crypto wire
Let me set the scene properly, because context is the only thing that separates analysis from noise.
Anthropic is the AI lab founded in 2021 by Dario Amodei and a group of researchers who left the frontier lab that shall go unnamed here โ the one whose model line everyone benchmarks against. Anthropic's entire public identity is built on one word: safety. Where its competitors sold capability first and worried about consequences later, Anthropic sold the consequences first. Its flagship product, Claude, is marketed less as the smartest model and more as the most trustworthy one. That is not a marketing accident. It is the company's core strategic bet, and it shapes everything the firm does.
The technical spine of that bet is something called Constitutional AI. Instead of paying thousands of human contractors to rank model outputs one by one โ the standard method, called reinforcement learning from human feedback, or RLHF โ Anthropic wrote down a set of written principles, a constitution, and trained the model to critique and revise its own answers against that document. The model becomes its own editor, guided by a rulebook that a human wrote.
This matters enormously for the story in front of us. Because the moment you accept that a written constitution governs a model's behavior, you have to answer a question that sounds philosophical but is actually brutally practical: who writes the constitution, and who gets to amend it?
That is not an AI question. It is a governance question. And governance is the one thing the crypto industry has spent the last ten years doing in public, badly, and at enormous cost to real people.
So when a crypto outlet reports that religious thinkers are now helping shape Claude's moral compass, I do not read it as an AI story that wandered onto a crypto wire by mistake. I read it as a governance story. The crypto wire published it because the crypto audience, whether it knows it or not, is the only mainstream audience that already understands what is at stake. We have watched this movie. We know how it ends when you get it wrong.
What the crypto industry learned about constitutions
Let me take you back, because the parallel is not decorative. It is the whole argument.
In 2016, a project called The DAO raised roughly one hundred and fifty million dollars in ether โ at the time, the largest crowdfunding event in history โ on the promise that it would be governed entirely by code and token-holder votes. No board. No executives. No lawyers. A constitution written in Solidity, enforced by a blockchain, amended by whoever held the most tokens.
Within weeks, an attacker drained roughly sixty million dollars by exploiting a flaw the constitution had not anticipated. The community faced a choice that no rulebook had addressed: let the code stand and accept the loss, or rewrite the rules and reverse the theft. They chose to rewrite. Ethereum forked. A minority refused the fork and kept the original chain alive as Ethereum Classic. The constitution had been amended by emergency decree, and the people who disagreed were told they could keep the old version if they wanted.
I want you to hold that sequence in your mind, because it is the exact sequence that Anthropic is now inviting. A written constitution. A gap the authors did not foresee. A decision about who has the authority to close the gap. A minority who feels the amendment was illegitimate.
The difference is that in 2016 the constitution governed money, and the amendment cost sixty million dollars and a permanent schism. In 2026, if the reports are accurate, the constitution governs the moral reasoning of a model that hundreds of millions of people will consult about their health, their finances, their relationships, and their politics. The blast radius is not sixty million dollars. It is the reasoning layer of an entire civilization.
And the crypto industry, which learned this the hard way, has a message for anyone paying attention: a constitution is only as legitimate as the process that produces it, and a process is only legitimate if the people bound by it believe they had a voice in writing it.
Everything else โ the tokenomics, the whitepapers, the vesting schedules โ is downstream of that single truth.
Why this is a governance story wearing a theology costume
The headline frames this as a story about religion. It is not. It is a story about representation, and representation is the oldest unsolved problem in decentralized systems.
Here is the mechanism, stated plainly. Anthropic's constitution โ whatever its actual text โ encodes a set of values. Those values determine what Claude will and will not say, how it hedges, what it refuses, what it treats as obvious and what it treats as contested. Right now, those values were written by a small group of people in San Francisco with a particular cultural formation. That formation is not neutral. It is a specific worldview โ techno-optimist, secular, liberal, Western โ dressed up as the default.
When critics accuse AI models of a Western secular bias, they are not making a philosophical complaint. They are making a governance complaint. They are saying: the constitution was drafted by an unrepresentative body, and I am bound by it anyway.
This is precisely the complaint that tore through the crypto governance debates of 2020 and 2021. Token-weighted voting was supposed to be the answer. Hold tokens, get votes. Except the people with the most tokens were the venture funds and the founding teams, which meant that governance was technically decentralized and practically plutocratic. The 1% voted, the 99% watched, and the constitution drifted toward the interests of whoever held the largest bags.
I remember auditing the token distribution of a mid-cap DeFi protocol during that period โ a project I will not name because the wound is still fresh for the people who held it โ and finding that the top four wallets controlled thirty-eight percent of the governance supply. The whitepaper said community-governed. The chain said four people. That gap between the document and the reality is the single most important thing I have learned to look for in my career, and it is the gap you should be looking for in the Anthropic story right now.
The religious thinkers, if they are real participants and not a press release, are an attempt to widen the drafting committee. They are an attempt to say: the constitution was written by more than the usual suspects. Whether that attempt is genuine or cosmetic is the entire question, and the thin reporting gives us no way to answer it.
The oracle problem, applied to values
I need to bring in something that will sound like a tangent and is not.
In DeFi, there is a component called an oracle. Its job is to tell smart contracts what is happening in the outside world โ the price of ether, the outcome of an election, the temperature in a warehouse. A smart contract cannot see. It can only execute. So it relies on an oracle to feed it truth.
The most important oracle in the industry is Chainlink, and Chainlink has a structural tension at its heart that I have written about for years. It markets itself as decentralized โ a network of independent node operators, no single point of failure. But in practice, the node set is small, the operators are professionalized, and the whole thing depends on a reputation-and-staking layer that is governed by a foundation. It is decentralization with a centralized load-bearing wall, painted over.
Why does this matter for Anthropic? Because a religious advisory council is an oracle for values. Its job is to tell the model's training process what the outside world believes โ what is sacred, what is forbidden, what is contested, what is settled. And like Chainlink, the council's legitimacy depends entirely on a question nobody has answered: who authorized these particular nodes to speak for the whole network?
If Anthropic invites three religious thinkers, which three? Chosen by whom? Speaking for which billions of believers? A Sunni scholar does not speak for Shia communities. A progressive rabbi does not speak for Orthodox congregations. A Buddhist monk does not speak for the secular majority of Japan. The moment you put a human being on a council and call them a representative, you have made a claim about representation that the represented have not consented to.
This is not a hypothetical. It is the exact failure mode that crippled the early oracle designs and forced the industry to invent staking, slashing, and reputation systems โ crude mechanisms for answering the question of who is allowed to speak truth. The crypto industry spent years and hundreds of millions of dollars learning that a trusted node is a contradiction in terms. Anthropic, if the reporting is right, is about to learn it for free.
The alignment tax, and who pays it
Here is where my financial engineering background earns its keep, and where I want to introduce a concept the AI safety world borrowed from economics without, I suspect, fully pricing it.
Economists talk about an alignment tax: the cost, in capability, of making a model behave well. Every constraint you add โ do not help with this, hedge on that, refuse the other โ costs you something. A model that refuses too much is less useful than a model that refuses too little. The tax is real, and it is paid in utility.
The crypto world pays an analogous tax every single block. It is called coordination overhead. Consensus is expensive. Every node that has to agree, every validator that has to vote, every governance proposal that has to clear a quorum โ that is latency, that is cost, that is friction. Bitcoin pays it in wasted electricity. Ethereum pays it in gas. DAOs pay it in the endless, exhausting forum threads that never reach a decision.
Now watch what happens when you multiply the drafting committee. A constitution written by one team is fast to amend and cheap to maintain. A constitution written by a committee of religious thinkers, secular ethicists, safety researchers, and product managers is slow, expensive, and prone to deadlock โ because the inputs genuinely conflict. Some traditions hold that a human life begins at conception. Others do not. Some hold that certain relationships are sacred. Others hold that the same relationships are forbidden. You cannot average these positions into a single rule without betraying someone.
So the model has three options, and each one carries a cost.
It can weight the inputs โ privilege some traditions over others โ which is a quiet way of saying that some believers matter more than others, and which will be discovered eventually.
It can hedge โ refuse to engage on contested questions โ which inflates the alignment tax until the model becomes useless precisely where people need it most: medical decisions, legal questions, moral dilemmas.
Or it can fragment โ different answers for different contexts โ which is coherent until a user asks the same question twice and gets two different moral universes.
None of these is free. The crypto industry learned this the hard way when it tried to make on-chain governance respond to every stakeholder. The protocols that survived were the ones that accepted a narrower, more honest mandate and stopped pretending to represent everyone. The ones that tried to represent everyone became paralyzed, and then they became irrelevant.
The alignment tax is not a technical footnote. It is the price of legitimacy, and somebody always pays it โ the question is only whether the bill lands on the users or on the model's capability.
The moat nobody wants to talk about
Now let me do what I actually do for a living, which is look at where the money is hiding.
There is a temptation, when a story like this breaks, to file it under reputation and move on. Reputation is soft. It does not show up in a valuation model. But I have learned, over twenty-one years, that the softest assets are often the hardest moats, and the Anthropic story is a moat story in disguise.
Think about what happened to Binance. In 2023, the exchange paid a fine of four point three billion dollars โ a number so large it read like a typo โ and settled a raft of criminal and civil charges. Conventional wisdom said the fine would cripple it. The opposite happened. The fine became the moat. Because once you have paid the price of admission, you hold a license that no newcomer can afford to buy. The entry ticket is now four billion dollars. The market consolidated around the survivors, and the survivors got stronger.
I wrote about that dynamic at the time and caught a lot of pushback for it. People wanted Binance punished, not entrenching. But markets do not care what people want. They care about barriers to entry. And a regulatory settlement, however painful, is the most durable barrier to entry ever invented, because it is written into law and enforced by the state.
Now translate that logic to AI ethics. If Anthropic builds a genuine multi-stakeholder values-governance process โ religious advisors, ethics reviews, auditable constitution amendments โ it is not just polishing its brand. It is buying the license that will eventually be required to sell AI into regulated markets. Financial services. Healthcare. Government. Every one of those sectors has a procurement committee that will, within a few years, ask a simple question: show me your value-alignment governance. And the labs that built it early will have an answer. The labs that did not will be paying consultants to fake one.
The entry ticket to enterprise AI is being priced right now, in real time, and the currency is not compute. It is governance. Anthropic is not buying a reputation. It is buying a license, and licenses are the deepest moats in any regulated industry.
The crypto angle that actually matters
I would be a poor exchange market lead if I ignored the assets. Let me be precise, because this is where a lot of retail money is going to get hurt if it reads the wrong signal.
There is a sector of the crypto market that has spent the last three years positioning itself as the decentralized answer to exactly this problem. Decentralized AI networks โ projects that tokenize model training, inference, or data โ market themselves on a single promise: no single company controls the values. Bittensor, Fetch, Render, and a dozen smaller names all sell a version of the same pitch. Governance by token. Alignment by market. No San Francisco committee deciding what is sacred.
If the Anthropic story is real and durable, it is a validation event for that thesis and a threat to it at the same time. It is a validation because it proves that value-alignment governance is now a first-order competitive concern โ the kind of thing serious labs spend real money on. It is a threat because it demonstrates that the centralized players can simply buy legitimacy by convening the right committee, which is a far cheaper path than rebuilding the entire stack on-chain.
I have watched this pattern before. In 2021, every layer-two scaling project promised that decentralization would solve the trust problem. Then the centralized exchanges just added more insurance funds and better cold storage, and most users never left. Convenience beats ideology, every single time, and it does so in bear markets fastest of all, when people are too tired and too poor to care about architecture.
So here is my read on the AI-token sector in this context. The narrative tailwind is real but thin. Any project whose entire pitch is we are the decentralized alternative to Anthropic just got a competitor that is willing to co-opt its best argument. If you are holding those tokens, the question you need to ask is not does this validate the thesis. It is does this validate my specific project, or does it validate the category while the centralized labs eat the category's lunch. Those are very different questions, and conflating them is how people hold through a drawdown that never recovers.
What the reporting refuses to say
I want to spend real time here, because this is the part that separates analysis from press release recitation, and because the silence in the source material is itself a signal.
Read the original item the way I read a token whitepaper. Ask what it tells you and what it carefully does not.
It tells you: religious thinkers are involved. It does not tell you how. Is this a formal advisory board with a charter and meeting minutes? A one-time consultation? A handful of interviews feeding into a training dataset? The distinction is not semantic. A formal board with decision rights changes model behavior. A one-time consultation changes a slide deck.
It tells you: moral compass. It does not tell you whether that phrase maps to anything in Anthropic's actual technical documentation. Anthropic does not use that language. It uses words like values, constitution, character training. When a journalist uses a metaphor instead of a term of art, it usually means the journalist did not have access to the technical detail โ which means the mechanism is unknown even to the reporter.
It tells you: religious thinkers, plural. It does not tell you which traditions. This is the most consequential omission. If the council is three flavors of Western Christianity, the gesture is cosmetic and the secular-bias critique stands untouched. If it genuinely spans Islam, Buddhism, Hinduism, Judaism, and Christianity, the conflict-management problem becomes severe and the alignment tax spikes. The composition of the council is the entire story, and it is the one fact the reporting withholds.
And it tells you none of the following: whether the advisors have veto power or advisory power only; whether their input is auditable; whether Claude's behavior changed measurably; whether there is an independent third party checking any of this; whether the same courtesy was extended to non-Western, non-Abrahamic traditions; and whether any of it is contractually binding on the company.
That is a long list of absences for a four-sentence story. In my line of work, a list of absences that long is not an accident. It is a shape. And the shape is the shape of a narrative that has been polished for public consumption while the mechanism stays in a drawer.
The contrarian angle: ethics as the newest moat, and the newest washing
Here is the reading I have not seen anyone publish, and I think it is the correct one.
The religious-advisor story is not primarily about religion, and it is not primarily about safety. It is about positioning for the regulatory endgame, and it arrives at the exact moment the regulatory endgame is being written.
The European Union's AI Act is phasing in. The NIST AI Risk Management Framework is being operationalized. ISO/IEC 42001, the AI management-system standard, is being adopted by enterprises that need a box to tick. Every one of these frameworks rewards documented multi-stakeholder governance. None of them specifies who the stakeholders must be. That gap is the opportunity, and Anthropic is filling it before anyone else does.
When the auditors come โ and they will come โ the lab that can say we convened religious, ethical, and technical advisors, we documented the process, we can show the amendments, will pass. The lab that cannot will scramble to retrofit. And retrofitting governance under audit pressure is expensive, slow, and embarrassing.
So the move is not charity. It is procurement strategy. It is the AI equivalent of buying the license before the license becomes mandatory.
But here is the dark side, and I will not soften it. There is a real risk that this is ethics washing โ the use of a diverse-looking process to launder a set of values that remains exactly as narrow as it always was. The tell is power, not presence. If the religious thinkers can point to a specific output that changed because of their input, the process is real. If they cannot, they are decoration, and decoration is a lie told politely.
The crypto industry has its own version of this. Every protocol has a community call and a governance forum, and on most of them, the founders decide and the community applauds. I have sat through dozens of those calls. The token holders ask questions. The founders answer vaguely. The proposal passes. Nothing changes. The governance was real in the sense that the meeting happened, and fake in the sense that the decision was made before the meeting began.
I am not saying Anthropic is doing that. I am saying the reporting gives me no way to tell, and in the absence of evidence, the base rate โ the historical frequency of this pattern across industries โ is not reassuring.
The secular backlash nobody is pricing
There is one more force in this story that the reporting ignores entirely, and it is the one most likely to bite.
A large fraction of the people who use AI tools are secular, and a meaningful fraction are actively hostile to organized religion. They are not a fringe. They are the core user base of every major AI product: engineers, researchers, writers, analysts โ disproportionately non-religious, disproportionately urban, disproportionately the exact demographic that adopted crypto early and then abandoned it when the culture turned.
When you announce that religious thinkers are shaping the model's moral compass, you are telling that demographic something specific: the values baked into your tool are now partly authored by institutions you may not trust. For some users, that is a comfort. For others, it is a betrayal. And the ones who feel betrayed are, historically, the loudest and the most likely to churn.
The crypto parallel is exact. When Ethereum forked to reverse The DAO hack, it made a values choice โ code is not law, community comes first โ and a minority of purists left forever. That minority never came back. They built Ethereum Classic, and they still write angry posts about it, and the chain is still alive, and the schism is permanent. A values decision, once made, does not heal. It hardens into tribes.
Anthropic is walking toward the same kind of decision. The moment it signals that religious tradition has a seat at the table, it hardens one tribe of users and alienates another. And in a market where the product is nearly interchangeable with two competitors, tribal loyalty is the only thing that keeps users from leaving.
There is also a jurisdictional problem that the reporting completely misses. France is a strictly secular republic. Its constitutional principle of laรฏcitรฉ is not a preference; it is law. A model whose values were shaped by religious advisors is, to a French regulator, not more legitimate โ it is less. The same governance move that earns points with a European AI Act auditor could cost points with a French data-protection authority. There is no universally correct answer here. That is the point. Value alignment is not a problem you solve. It is a problem you manage, forever, in public, with everyone watching.
What I would actually watch
I want to close with something usable, because analysis that does not change your behavior is just entertainment, and I do not write entertainment in a bear market.
Watch for disclosure. If Anthropic publishes the composition of the council, the terms of their engagement, and at least one documented change to the model's behavior that resulted from their input, the process is real and the moat is being built. If silence persists past the next few months, treat the whole thing as brand maintenance and price it accordingly โ which is to say, price it at zero.
Watch the model's behavior on contested questions. Alignment leaves fingerprints. If Claude's responses on medical ethics, end-of-life decisions, or family law shift in a way that tracks a particular tradition's doctrine, that is evidence of real influence. If the outputs stay exactly as they were, the advisors were decoration.
Watch the competitors. If the other frontier labs launch their own multi-stakeholder value-governance programs within a year, this was a standard-setting race and Anthropic won the first lap. If nobody follows, it was a solo press play and the moat is imaginary. In crypto, the standard-setting races are visible from the first proposal. Watch whether a second proposal appears.
Watch the decentralized AI tokens โ but carefully. The category gets a narrative boost from this story. Individual projects may not. Do not confuse the validation of a thesis with the validation of your specific bag. I have watched too many people ride a category up and a project down.
And watch the regulatory calendars. The AI Act's high-risk obligations are phasing in over the next several years. The labs that show up to that deadline with a documented governance process will win enterprise contracts by default. The labs that show up empty-handed will be buying legitimacy at whatever price the market sets โ and the market, as Binance taught us, sets a brutal price for late entry.
The quiet, and what it is actually telling us
Let me return to the silence I started with, because it is the most honest thing in this entire story.
We do not know what Anthropic did. We know a crypto outlet said something happened, in four sentences, without a mechanism, without a name, without a date. Everything I have written above is a map of the terrain around that absence โ the governance terrain that the crypto industry mapped first, at the cost of billions, and that the AI industry is about to cross with far more at stake.
The lesson my industry learned the hard way is the one I want to leave you with. A constitution is not legitimate because it is written. It is legitimate because the people bound by it believe they helped write it, and because they can see, at any moment, whether the rules they agreed to are the rules being enforced. The moment that visibility disappears โ the moment the process becomes a slide deck instead of a decision โ the constitution becomes a costume, and costumes do not hold power when the weather turns.
Anthropic may be building something real. It may be buying a license it will genuinely need. It may be doing the most consequential governance experiment in the history of technology, and it may deserve the benefit of the doubt that its own reporting fails to earn.
Or it may be doing what every institution does when it is criticized for a narrow worldview: convening a committee, taking a photograph, and changing nothing.
The silence will not tell us which. The behavior will. And behavior, unlike press releases, cannot be laundered. It leaves fingerprints on the outputs, on the audits, and on the users who stay or leave. Watch those, and you will know more than the headline ever intended to reveal.
The cheetah does not chase the noise. It watches the grass move, and waits for the thing that made it move to show itself. The grass is moving right now. What made it move has not yet appeared. Do not trade on the ripple. Trade on the animal.
And in a bear market, that discipline is not optional. It is the difference between surviving the cycle and becoming a footnote in someone else's.