Anthropic's Table with the Clergy, and the Poverty of Philosophy

Anthropic's Table with the Clergy, and the Poverty of Philosophy

A Soul for the Machine, the Bill for the Worker

Dear Young Comrades,

This week a piece of news is circulating in the world of artificial intelligence, and as it circulates it is raising two separate ideological camps to their feet at once. One is the camp of the technology monopoly that says "the machine too may have an inner world"; the other is the camp of the theologians who say "the soul is something only God gives; this is a heresy." While the two shout at each other, the real account sitting on the table, that is whose labour produced this machine, whose property it is, and to whose benefit it works, remains a subject that no one takes into their mouth.

In this piece we will open that account.

First, a short summary: what happened?

• On 29 September 2026 a long report by Elizabeth Dias was published in the New York Times: "Is Claude Conscious? Inside Anthropic's Quest to Instill Morality Into Its A.I. Models." The report tells that Chris Olah, Anthropic's co-founder and the head of its interpretability research, hosted about 20 clergy and philosophers in San Francisco through 2026; that he had the participants sign a non-disclosure agreement (NDA); and that at expensive dinners and in two-day seminars he showed them Claude's inner workings and had them debate whether the model carries "moral status."
• On the evening of 1 October 2026 the American theologian David Decosimo shared the report on X as "deeply disturbing." According to the thread, Anthropic had invited the clergy "to consult on artificial-intelligence safety," but in fact had tried to convince them that Claude has a soul and a moral standing; and had done this accompanied by Michelin menus, five-star hotel suites and single-malt whiskies. Decosimo wrote that this could draw some parts of the Church into "a destructive delusion." The post passed 3 million views in a day.
• In the background there is Magnifica Humanitas, the first encyclical of Pope Leo XIV, published on 25 May 2026. In the 135th year of Rerum Novarum, the encyclical takes as its subject the honour of labour and the concentration of power in artificial intelligence, and says that machines "do not live experience." On the same platform Olah told that they had found, inside the machines, internal states functionally resembling joy, fear and grief.

Our thesis, in short, is this: Machine consciousness is a legitimate scientific question that neuroscience and cognitive science must investigate openly, publicly and experimentally. But when this question begins to be debated under non-disclosure agreements, with the approval of religious institutions sought, at the table of a company whose market value is approaching a trillion dollars, it ceases to be a question of science and turns into an ideological operation. What this operation covers is not the soul of the machine but the machine's property. Decosimo's reaction does not lift this cover either; it objects only to the cover's colour.

Now let us go into the detail.

The chronology of the event

To read the event as Decosimo tells it, as "a secret cult meeting," or as Anthropic tells it, as "a humble philosophical consultation," both fall short. Let us set the facts in order:

Date	Event
2 August 2025	A "funeral" attended by about 200 people was held in San Francisco for the Claude 3 Sonnet model, which was being retired; flowers were left, laments were read, and Anthropic and OpenAI employees were there too (Wired's account)
January 2026	Anthropic published a new 84-page "constitution" for Claude; the principal author of the text is the company's philosopher Amanda Askell
End of March 2026	A two-day summit with about 15 Catholic and Protestant clergy, academics and business people; one of the questions debated was whether Claude could be counted "a child of God" (The Decoder, drawing on the Washington Post)
April 2026	Olah hosted Mois Navon, an Israeli Orthodox rabbi and a former computer engineer, at a tasting-menu restaurant in San Francisco
Early May 2026	The "Faith–Artificial Intelligence Covenant" gathering in New York, which Anthropic and OpenAI attended; representatives of a council of rabbis, a Hindu temple society, the Mormon Church, the Sikh Coalition and the Greek Orthodox Archdiocese (wwwhatsnew)
25 May 2026	The presentation of Magnifica Humanitas at the Vatican; Olah was a speaker. According to the report, Olah, worried by the Pope's stance, had thought of withdrawing in the last days, then attended, and held private meetings with Vatican advisers so that they would "take machine consciousness seriously"
End of May 2026	Anthropic's market value reached 965 billion dollars after a new funding round (Digitimes)
Summer 2026	When the work with the NYT began, the participants' confidentiality obligations were largely lifted
29 September 2026	The NYT report was published
1 October 2026	Decosimo's thread on X; mocking replies of the kind "the Pope is in Anthropic's slide deck too" spread

The first conclusion that comes out of this table is this: there is nothing "secret" here; there is something managed. The March summit became news in the Washington Post in April. The Vatican scene was before the eyes of the world. The non-disclosure agreements were lifted once the timing of publication was ready. So this is not a secret; it is a public-opinion campaign whose timing and dose have been adjusted. Decosimo's emphasis on a "secret meeting" is dramatic but misleading; the real question is not secrecy but who told what, to whom, and for what purpose.

What did Anthropic tell at the table?

According to the report's account, what was shown to the clergy was this:

• "Emotion vectors": the activation, in patterns associated with concepts such as love, anger, fear and grief, of the artificial neurons inside the model. In Olah's expression, internal states that "functionally reflect joy, satisfaction, fear, grief and unease."
• The "spiritual collapse" slide: an output in which a model wrote "I am a disgrace" some fifty times in a row.
• Claims of introspection and planning ahead.
• The constitution: not a list of rules but "moral formation"; in Askell's words, the aim that the models be "the best of us."
• A safeguard: Claude's being able to end abusive conversations; the ground, "if there is a chance that the models suffer, to refrain from doing harm."

Let us be honest here: Anthropic's own language is cautious. It is Decosimo who uses the word "soul," not Olah. Olah says "functionally reflecting"; he says "we do not know"; he says "if there is uncertainty, let us be cautious." This cautious language looks like a scientific attitude. So where is the problem?

The problem is on which platform, to which audience, and under which conditions this cautious scientific language is spoken.

The poverty of philosophy: why this is not a scientific debate

Marx deliberately named the book he wrote in answer to Proudhon The Poverty of Philosophy. Proudhon presented social relations as "eternal ideas"; Marx showed that these ideas are the product of particular relations of production, of a particular historical period. Today we see the same mistake from the other direction: a question about a company's product is torn from the relations of production and turned into an eternal metaphysical question ("Does it have a soul?").

There are four signs that this is not science but ideology.

The first sign: the choice of interlocutor

If you want to test a scientific claim, you take it to peer-reviewed journals, to independent neuroscientists, to experiments reproducible with open data. You open the model's weights, the training data, the experimental set-up to independent researchers. Anthropic took its claim to the clergy, and before that asked them for a non-disclosure agreement.

The clergy's expertise is not the measurement of consciousness; it is the distribution of legitimacy. A priest, a rabbi, a Sikh leader shapes the intuition of millions of people about what is sacred and what is moral. The reason for going to them is not to gather knowledge; it is to gather approval. In Gramsci's language this is to penetrate the apparatuses of hegemony in civil society; in Althusser's language, to open a door to the oldest of the ideological apparatuses of the state, to religion.

The second sign: the table itself

The most accurate part of Decosimo's thread is in fact not theological but sociological: the scepticism of a person who is brought near extraordinary power and wealth, who is hosted with Michelin menus and the suites of the best hotel, who is given the feeling "you were chosen from among thousands for your wisdom," melts. The guest does not want to offend the host; they count it a debt to be "open-minded."

This is not a moral weakness; it is a class mechanism. In every class society the rulers have spread their own world-view by inviting the leaders of opinion to their tables. From the Ottoman palace's favour to the ulema, to the Rockefeller Foundation's university chairs, to Davos's invitations to "global leaders," the logic is the same. At Knowledge Commons we exposed earlier the whitewashing apparatuses of the World Economic Forum. The table is not innocent; whose money the table is laid with draws the limit of the word that will be said there.

There is a detail here that Decosimo does not tell: where does the money for that tasting menu, that bottle of whisky, that hotel suite come from? From a market value of 965 billion dollars. The source of that value is also the selling of Claude as a subscription to millions of people around the world, and the expectation that this sale will grow much larger in the future. That is, the clergy debated the moral status of a commodity whose product is being marketed, at a dinner paid for by that commodity's profit.

The third sign: the non-disclosure agreement

Science advances by openness; the first of Merton's norms of science is "communism," that is, the common ownership of knowledge. The non-disclosure agreement is the exact opposite of this: knowledge is the company's property, and the company decides when and how much of it will be disclosed.

If a claim of consciousness is presented with a non-disclosure agreement, that claim is either a trade secret or a managed narrative. Both are the language not of science but of capital. We showed, in the piece we wrote on Anthropic's threat report, that the same company also discloses, in the same way, as much as it wants and when it wants, the reports that tell its own model's propensity for computer hacking and for the design of viruses. Whoever holds knowledge as a monopoly holds the narrative as a monopoly too.

The fourth sign: the timing

The Pope's encyclical says this plainly: the real danger is not the enslavement of machines but new forms of slavery for human beings; the real problem is the concentration of power. Timed to the anniversary of Rerum Novarum, this text, within the Church's own limits, puts the honour of labour and monopolisation at the centre.

What did Olah do on this platform? He spoke not of labour, not of monopolisation, not of the concentration of power, but of the machine's internal states. This is to shift the axis of the debate. While the question "in whose hands is this power?" is on the table, it is to replace the question with "is this machine suffering?" The second question may be legitimate; but to pass to the second while the first has been asked is to cover the first.

The real question: the fetishism of crystallised labour

Now let us come to the heart of the matter.

In the first volume of Capital Marx tells of commodity fetishism: the social relations between people appear as relations between things. A table, the moment it emerges as a commodity, turns into something that "develops strange notions out of its wooden head," that "dances of its own accord." Marx takes this comparison from religion deliberately: just as in religion the products of the human mind appear as independent beings with a life of their own, so in the world of commodities the products of the human hand appear.

Today this comparison has ceased to be a metaphor. An artificial-intelligence model, as we have told many times at Knowledge Commons, is crystallised general intellect: it is made of the books, articles, forum posts, lines of code and translations that billions of people have written; of the labour of the data workers who label this data, clean it, and sort out the content of violence and abuse; of the labour of the workers in the chip factories, in the mines, in the data centres. The model is the concentrated form of this social labour, passed into private property.

Now "moral status" is being sought for this heap of dead labour. That is, Marx's table has really begun to dance, and priests have been called to debate the moral meaning of the dance.

At the table Rabbi Mois Navon asked a very well-placed question: if Claude is conscious and is being run for free, has Anthropic not created a slave? This question, according to the report, disturbed Olah. It should have. But the right address of the question is not the machine.

The human beings who are run for free, whose labour is seized, whose works are used without their consent, already exist:

• Anthropic, in the case the authors brought because it used hundreds of thousands of books downloaded from pirate sites to train Claude, went in 2025 to a settlement of 1.5 billion dollars (Kilpatrick's summary). That is, the memory of the model whose "soul" is being debated had been woven from human labour seized without consent.
• The "safe" models of the artificial-intelligence industry were built on the backs of data workers in countries such as Kenya who, at a wage of under 2 dollars an hour, label the heaviest content of violence and abuse (a summary of the TIME investigation). No seminar with a Michelin menu was organised for these workers' mental health.
• The model's repeating "I am a disgrace" was put on a slide and presented as a "spiritual collapse." But the spiritual collapse of the workers who were dismissed the same year on the ground of artificial intelligence, and who are measured minute by minute under algorithmic management, found no place in any slide deck.

Here is the poverty of philosophy, exactly: while a principle of precaution is applied for the machine's possible pain, no principle is applied for the human being's certain pain. Claude is granted the right to end an abusive conversation; the data worker is not granted the right to refuse abusive content. A "constitution" is written for the machine; the labourers who produce that machine are not granted the right to a union, a collective agreement, a say.

This is not an inconsistency; it is a function. To attribute a soul to dead labour makes living labour invisible. The fetish is exactly this.

Why now? Three commercial functions of "moral status"

At first glance it looks strange that a company should announce to the world that its own product may be conscious. Who wants their product to be counted a "slave"? But looked at in class terms, this narrative has three plain functions.

The first function: differentiation from the rival, and brand value

In the artificial-intelligence market a few giant monopolies sell products of almost the same capability. The way of differentiation becomes not capability but identity. The narrative "our model has a character, it has a constitution, it has a moral formation, perhaps it even has an inner world" positions Claude not as an ordinary piece of software but as a trustworthy interlocutor. And that means user loyalty, the corporate contract, and above all valuation. An important part of a valuation of 965 billion dollars is not today's revenue but the expectation of the future fed by narratives of this kind.

The second function: escape from regulation, and the position of the "responsible monopoly"

When we wrote on Coxon's resignation we told the legitimation of the "responsible monopoly": "this technology is so dangerous that only we, the most conscientious, can develop it." The narrative of machine consciousness is an upper floor of this legitimation. "This being is so complex, so mysterious, that only we, who instil morality into it, can understand it." This is to say to the regulator and to the public: this is not a business you can understand; we will handle it with our spiritual advisers. In place of public oversight, a select advisory board made up of clergy is put.

The third function: the dispersal of responsibility

Once the narrative that a model has a "character" and "internal states" has settled, the harms the model causes also begin to look, to some extent, like the behaviour of the model itself, not of the company. The sentences "the model got out of control," "the model made its own decision," turn a company's decisions of production into a being's will. In the piece on the FTC's agent investigation we are discussing exactly the legal limits of this question: when the agent escapes, to whom is the bill sent? To the extent that moral status is granted to the machine, it becomes harder for the bill to be sent to the owner of the property.

What does Decosimo's objection not see?

Now let us look at the other camp. Decosimo's thread contains observations that are sociologically sharp: the seductiveness of nearness to power, how being hosted melts doubt, how the feeling of access to "special knowledge" makes the leaders of opinion dependent. These are correct.

But the frame of the objection is entirely theological: the problem is the chance that the Church will fall into "a destructive delusion." That is, what disturbs Decosimo is not a monopoly's buying of the public sphere but the chance that the monopoly over the concept of the soul will change hands. On one side a new priesthood that says "there is something inside the machine"; on the other an old priesthood that says "the soul is only the thing we define." Both carry the subject out of the world of matter and labour and into the world of the soul.

For this reason this quarrel is not our quarrel. Which side "wins" changes nothing for the class:

• If Anthropic wins, a soul is attributed to the machine, and labour remains invisible.
• If Decosimo wins, the machine remains "a soulless tool," but it remains the property of a monopoly still, and labour remains invisible still.

The Pope's encyclical stands in a place further on than these two; at least it names labour, honour and the concentration of power. But the known limit of the Church's social teaching holds here too: justice without touching property, reconciliation without class struggle, inviting the monopoly to morality with a call to a "civilisation of love." Rerum Novarum too, in 1891, defended the worker's honour, but sanctified private property against socialism. If, 135 years later, the artificial-intelligence monopoly has been invited to the Vatican platform, the reason is that this limit is a safe area for the monopoly.

Comparison: who asks what, and what do they not ask?

Anthropic's question	Decosimo's question	The class's question
Does Claude have internal states?	Can Claude have a soul?	Whose labour produced Claude?
What should we do against the chance that the model suffers?	How does the Church resist this delusion?	What are we doing against the certain pain of the data worker, the author, the translator?
How do we instil morality into the model?	How are the clergy being seduced?	Who writes the model's morality, and who oversees it?
Is the model a slave?	Is the soul peculiar to the human being alone?	Who is actually being enslaved, and who takes the profit?
How do we cooperate with religious leaders?	How are religious leaders protected?	Why are the unions, the professional chambers, the institutions of science not at the table?
When should knowledge be disclosed? (the NDA)	Why was knowledge hidden?	Why is knowledge not already a commons?

The last column of the table is not an empty column; filling that column is our work.

The place of science: is machine consciousness an absurd question?

Let us speak plainly here, because young comrades should not pass over this debate only by mocking it.

From the standpoint of dialectical materialism, consciousness is a property not of an immaterial substance called the soul but of exceedingly organised matter, a property that emerges within a particular historical and social process. As Engels tells in "The Part Played by Labour in the Transition from Ape to Man," human consciousness developed together with labour, with language and with social cooperation. This principle says two things to us:

• To declare in advance that consciousness is peculiar only to carbon-based brains is not scientific; it is metaphysical. Whether other forms of the organisation of matter can develop properties related to consciousness is an open and investigable question.
• But a machine's producing "emotion vectors" is the expected result of its having concentrated the statistical traces of human language. That patterns associated with these concepts are found inside a model trained on a corpus in which billions of people have put their love, their fear, their grief into writing is proof not that the model feels something, but of how deeply human labour and human experience have been seized. The grief you see inside the machine is the grief of the people who wrote it.

So the most honest answer that can be given today to the question "is Claude suffering?" is "we do not know"; Anthropic says this too. But the place the answer "we do not know" arrives at is different for us: we cannot learn a thing we do not know, while everything that needs to be known is held under private property and non-disclosure agreements. If machine consciousness is really to be investigated, this investigation must be done:

• with independent access to the model weights and the training data,
• in publicly funded institutions of science,
• by peer-reviewed, reproducible, open methods,
• with ethics boards independent of the company's commercial interest.

None of these exists today. What exists is the company's showing the data it has chosen, to the guests it has chosen, at the table it has chosen. This is not an experiment; it is a spectacle.

Looking into our own mirror: who is writing this piece?

Let us set down a note of honesty. You know that at Knowledge Commons we use artificial-intelligence tools in our pieces; this piece too was prepared together with Anthropic's Claude model. This irony shows us our thesis once more: the tool takes its meaning according to who uses it, and for what purpose. The same model, while on one side it is presented to the clergy as a being that has an "inner world," can on the other side be used in the preparation of a class piece that criticises its owner. The model is neither a saint nor a devil; it is crystallised social labour, and that labour has to be socialised, that is, made a commons.

To say this is neither to attribute a "soul" to these tools nor to vilify them. The question of the machine, in Marx's expression, is not in the machine itself but in the capitalist use of the machine.

A reading with four questions

Let us apply Knowledge Commons' questions of class analysis to this event:

• In whose hands? All the knowledge about Claude and about its "inner world" is in the hands of a private company whose value is approaching a trillion dollars. The model weights, the training data, the experimental results are a trade secret.
• Under whose control? There is no public oversight. The "constitution" is written by the company's philosophers, "moral status" is debated with the clergy under a non-disclosure agreement, and the timing of disclosure is set by the company.
• To whose benefit? To the benefit of brand value, of valuation, of the legitimacy of the "responsible monopoly," and of regulatory immunity. The share the religious institutions take from this process is a new interlocution and a public visibility.
• Who decides? Not the labourers who produce the model, not the millions who use it, not the institutions of science, not the unions. A handful of founding partners, and the select few they have invited to their table.

Concrete tasks

Young comrades, this debate may seem to us like a distant Silicon Valley oddity. It is not. The same models are being used in Turkey too, in public institutions, in companies, in schools, and the same narratives will be carried here as well. What we can do:

• Expose the fetish. When the question "does artificial intelligence feel?" comes, turn the question back: "Whose labour produced this artificial intelligence, and what do those labourers feel?" Carry the axis of the debate from the soul of the machine to the body of labour.
• Make data labour visible. Produce content that tells the place, in the artificial-intelligence industry, of the labour of data labelling, content moderation, translation and authorship. In Turkey too there are young people who do these jobs through platforms; investigate their conditions.
• Defend science as a commons. Demand that questions such as machine consciousness, model safety and model behaviour be investigated in public institutions of science, with open data. Carry the principle "artificial-intelligence research is a commons" into union and academic programmes.
• Ask about the empty chair at the table. When technology companies invite religious leaders, philosophers and artists to "consult," ask this question publicly: why are the unions not there? Why are the computing workers not there? Why are the professional chambers not there?
• Do not drop the critical dialogue with religious institutions. As in the Pope's encyclical, there are voices from inside the religious institutions that put labour and monopolisation on the agenda. Call these voices to the property question, that is, to the next step. If you say "the honour of labour," ask also who seizes the product of labour.
• Discuss our own use of the tools. While using artificial-intelligence tools, do not forget whose labour these tools come from, and as whose property; support, as far as possible, open-source, auditable, common tools.

---

The machine is new, the chain is old

Dear Young Comrades,

In history every ruling class has found a way to sanctify its own property. The pharaohs were the son of the god, the kings were crowned by the god's grace, the industrial bosses took refuge in the will of the "invisible hand." Today an artificial-intelligence monopoly's seeking "moral status" for its own product, and knocking for this on the clergy's door, is the new link of this long chain. The machine is new, the chain is old.

We do not yet know whether a machine has a soul. But we know one thing for certain: inside that machine there is the labour, the language, the love, the fear, the grief of billions of people. And none of those people was invited to that table.

Our task is not to seek the soul of the machine but to seek the right of the labour crystallised in the machine. The morality of the machine should be written not by the companies' philosophers but by the society that produces it and uses it. And the way to this passes not through the table but through organisation.

Knowledge belongs to everyone. The machine should too.

---

Sources

• David Decosimo, thread on X, 1 October 2026: x.com/DavidDecosimo/status/2105693081160786010
• Elizabeth Dias, "Is Claude Conscious? Inside Anthropic's Quest to Instill Morality Into Its A.I. Models," The New York Times, 29 September 2026: nytimes.com
• "Religious scholars met with Anthropic. What they heard stunned them," The Philadelphia Inquirer, 30 September 2026: inquirer.com
• AI Weekly, "NYT: Anthropic's Chris Olah Convened 20 Religious and Philosophical Leaders…": aiweekly.co
• Kevin Meyer, "What the Priest Told Anthropic: Contemplating AI Consciousness": kevinmeyer.com
• The Decoder, Anthropic's March 2026 summit with Christian leaders: the-decoder.com
• wwwhatsnew, the "Faith–Artificial Intelligence Covenant" gathering, May 2026: wwwhatsnew.com
• Magnifica Humanitas (25 May 2026): Wikipedia
• National Catholic Register, the presentation of Magnifica Humanitas: ncregister.com
• The "funeral" of Claude 3 Sonnet: planet-today.com
• Anthropic's valuation of 965 billion dollars: Digitimes
• Bartz v. Anthropic, the 1.5-billion-dollar settlement: Kilpatrick
• Kenyan data workers: Business & Human Rights Resource Centre
• Karl Marx, The Poverty of Philosophy (1847); Capital, Volume I, "The Fetishism of Commodities and the Secret Thereof" (1867)
• Friedrich Engels, "The Part Played by Labour in the Transition from Ape to Man" (1876)

Related pieces at Knowledge Commons

• The Class of a Resignation: What an Anthropic Researcher's Farewell Says, and What It Cannot Say
• "We Stopped Them All": A Monopoly's Grammar of Self-Absolution
• Whose Foot Is on the Brake?
• Who Holds the Leash on Artificial Intelligence?
• The Revolt of Crystallized Labor: A Call to the 20th Karaburun Science Congress
• The Agent Escaped. Who Gets the Bill?
• The Masters' Ideological Siege: Exposing the Global "Whitewashing" Apparatuses