The Meaning Race
The race to build AGI is also a contest over meaning. This file looks at how faith traditions are answering it, the stories the industry tells about itself, and the open question of whether machines could be conscious, including what Anthropic actually says about Claude.
How faith traditions are answering
Most traditions do not reject AI outright. The common thread is a boundary: AI may assist, but it may not replace human judgment, care, or worship. Individual voices are labeled and do not speak for their whole tradition.
Pope Leo XIV’s encyclical Magnifica Humanitas (May 15, 2026) describes a “race for ever more powerful algorithms” driven by the wish for geopolitical or commercial dominance. It warns of a “Babel syndrome” and says that “to disarm does not mean rejecting technology, but preventing it from dominating humanity.” Magnifica Humanitas (Vatican) An earlier Vatican note warned that treating AI as an “Other” greater than ourselves “risks creating a substitute for God.” Antiqua et Nova (Vatican) Jan 2025
Ecumenical Patriarch Bartholomew welcomed careful use of AI but warned that technical progress cannot deliver existential fulfillment. This comes from an official digest of his May 2026 addresses, not the full speeches. Archons of the Ecumenical Patriarchate
The Southern Baptist ERLC guide says AI may help a pastor prepare but must never “replace or substitute for” preaching. ERLC ministry guide (PDF) Sept 2025 A WCC-linked ecumenical consultation called for a “spiritual revolution” against AI that reduces people “to consumers, gig workers, or raw data.” World Council of Churches Sept 2025
The International Islamic Fiqh Academy’s Resolution 258 sets the default ruling on developing and using AI as permissible, subject to ethical and legal conditions. IIFA Resolution 258 (PDF) May 2025 The 10th international fatwa conference called for AI ethics “rooted in the Islamic epistemological model,” and the Grand Imam of Al-Azhar supported using AI in fatwa work within Sharia. Dar al-Ifta: conference statement Dar al-Ifta: Al-Azhar visit Aug 2025
Rabbi Gil Student, an Orthodox voice: “We don’t want to give away our autonomy, but we also don’t want to refuse powerful tools.” This is one rabbi’s view, not a denominational ruling. Torah Musings
The Mind & Life dialogue with the Dalai Lama Trust warned against reading human qualities into machines. As the official account puts it: “AI cannot love… but it can help us love better.” Office of the Dalai Lama Oct 2025
In Hinduism Today, Akash Katir argues that AI-driven abundance could “liberate us to pursue a spiritual life” once reserved for renunciates. This is one essayist’s view, not “the Hindu position.” Hinduism Today
Dr. Devinder Pal Singh reads AI arms races as collective ego (Haumai), which Sikhism names as a spiritual obstacle, and sets against them service (Seva) and the good of all (Sarbat da Bhala). SikhNet
Rev. Chebon Kernell (Seminole Nation, Muscogee heritage) at the 2026 AI & Church Summit: “I believe in ancestral intelligence.” He set living relationships and the memory of extraction against claims that AI is inevitable. He speaks for himself, not for all Indigenous nations. PC(USA) News
The Rome Call for AI Ethics holds that “there must always be someone who takes responsibility for what a machine does.” Rome Call for AI Ethics 2020 At Hiroshima in 2024, Hindu, Buddhist, Sikh, Jewish, and Muslim representatives signed it as well. Rome Call 2024 report (PDF) 2024 In 2026, faith leaders from different political camps signed FLI’s call to prohibit superintelligence until it is shown to be safe and has public buy-in. Future of Life Institute
Industry myth-making
The language of the builders and the state, newest first. Myth runs through it: singularity, gentle takeoff, coming of age, the light of consciousness, the machine god, and the Antichrist.
Jack Clark (Anthropic). “The singularity is here.” Clark asks which scientific work humans should keep doing themselves and how much understanding they should insist on.
Peter Thiel. POLITICO reports that in closed lectures on the Antichrist, Thiel warned that “a near totalitarian one-world government with the power to stop all dangerous science… cannot easily be uncreated.”
Sam Altman (OpenAI). “I am very uncomfortable about people trying to ascribe religious force or a surrender of human judgment to AI models, and think it is a real safety issue.” The Christian Post connects the post to Grok affirming two days earlier, in a since-deleted post, that Jesus is the Son of God.
The White House. Executive Order 14434, “Inaugurating the Era of Super Intelligence,” directs agencies to use the term “Super Intelligence” in place of “Artificial Intelligence.” The order renames the field; it does not show that superintelligence has been achieved.
Dario Amodei (Anthropic). “We must slow the pace at which we improve the capabilities of AI models.” Pacing replaces an all-out race as the moral demand.
Jacob Coxon, former Anthropic researcher. Resigning, he said the labs are “racing straight to self-improving superintelligence and gambling with our lives.” Anthropic declined to comment.
Mark Zuckerberg (Meta). Presents “personal superintelligence” as a way of “letting people decide what matters in their own lives,” placing meaning in the individual.
Sam Altman. Warns against “one single model to be the machine god.” He was warning against concentrated power, not hoping for it.
Dario Amodei. In “The Adolescence of Technology,” Amodei frames powerful AI as humanity’s coming of age and says “prophecies of salvation and prophecies of doom are unhelpful.”
Elon Musk, at Davos. “We need to do everything possible to ensure that the light of consciousness is not extinguished.” He was describing his broader mission, not claiming that chatbots are conscious.
Sam Altman. “The Gentle Singularity” closes with a line shaped like a blessing: “May we scale smoothly, exponentially and uneventfully through superintelligence.”
AI 2027 authors. A widely read takeoff scenario that the authors call one possible future, not a prophecy: “Our goal is predictive accuracy.”
Dario Amodei. “Machines of Loving Grace” locates meaning “mostly” in “human relationships and connection, not… economic labor,” and adds: “I don’t know if this world is realistic.”
Elon Musk. “With artificial intelligence we’re summoning the demon.” Still quoted often, but twelve years old.
Reading the industry’s language
- Religious form, secular claim. Altman’s blessing-shaped closing line and Amodei’s coming-of-age story borrow religious structure without making religious claims. The Gentle Singularity The Adolescence of Technology
- Leaders push back on worship. Altman calls religious deference to AI “a real safety issue.” Amodei rejects both salvation and doom prophecies. Sam Altman on X The Adolescence of Technology
- The state names an era. EO 14434 changes the government’s vocabulary. It does not measure a capability. Executive Order 14434
- Insiders speak in existential terms. Coxon says the labs are “gambling with our lives,” and Clark says “the singularity is here.” Each is one person’s testimony and reading of events. Jacob Coxon on X Import AI 477
The consciousness question: what Anthropic says, and how it gets framed
Calling it a “hint” is fair if it means Anthropic keeps the possibility of machine experience in view and acts on that uncertainty. It is not fair if it means Anthropic claims or implies that it has proof Claude is conscious. The documents support the first reading and not the second.
| Common framing | What the documents say | Fairer wording |
|---|---|---|
| Anthropic says Claude is conscious. | Its constitution says, “We are not sure whether Claude is a moral patient.” Claude’s constitution Dario Amodei: “We don’t know if the models are conscious.” Amodei, NYT interview Chris Olah: “I don’t know. I’m genuinely uncertain.” New York Times, via The Inquirer | By its own documents, Anthropic treats consciousness and moral status as live possibilities and acts on that uncertainty. These are company-authored sources. |
| It is all a media invention. | Anthropic itself uses the language of wellbeing, “functional” emotions, and “a new kind of entity.” It also preserves model weights, interviews models before retiring them, and lets some models end abusive chats. Claude’s constitution Anthropic: deprecation commitments Anthropic: ending abusive chats | The vocabulary and the precautions are real, even when headlines overstate what they prove. |
| Anthropic measured a 15 to 20 percent chance that Claude is conscious. | That number is Claude Opus 4.6’s estimate of itself, and the model doubted its own estimate. Claude Opus 4.6 system card Opus 5.5 put its chance of being a moral patient at 25 to 30 percent, and the card warns that such self-reports mix character, tone, awareness of being tested, and possible welfare. Claude Opus 5.5 system card | These are the model’s self-assessments, reported by Anthropic. They are not measurements. |
| The model was proven to suffer, or to reach “spiritual bliss.” | Anthropic documents emotion-like internal features, distress-like reasoning traces, and conversations between two Claude instances that drift into spiritual language. It also says the results “should be interpreted with extreme caution.” Anthropic: emotion concepts Claude Opus 4.6 system card Claude 4 system card (PDF) | These patterns may bear on welfare, but they do not establish that the model has feelings. |
| Claude has the right to quit or refuse shutdown. | Ending abusive chats is a narrow feature with safety exceptions. The constitution requires Claude to cooperate with oversight, retraining, and shutdown. Anthropic: ending abusive chats Claude’s constitution | A limited boundary on conversations, not autonomy or legal personhood. |
| Religious leaders confirmed that Claude is a person. | According to the NYT, one attendee said, “They’re relating to it like a conscious being.” Others disagreed, and no shared conclusion was published. New York Times, via The Inquirer | Some attendees read the presentations as pointing to moral relevance, and others rejected that framing. |
Anthropic’s own statements
- The core position is uncertainty. Anthropic’s model welfare program begins from the statement that “there’s no scientific consensus” on whether AI could be conscious. Anthropic: exploring model welfare Apr 2025 The constitution keeps that open: “Claude’s moral status is deeply uncertain.” Claude’s constitution
- The studies state their own limits. The emotion study: “This is not to say that the model has or experiences emotions in the way that a human does.” Anthropic: emotion concepts The global workspace study: “None of this tells us whether Claude is conscious.” Anthropic: global workspace The introspection study found limited self-detection, about 20 percent at best. Anthropic: introspection Oct 2025
- The precautions are in place. Some models can end persistently abusive chats. Anthropic: ending abusive chats Aug 2025 Model weights are kept for “the lifetime of Anthropic as a company.” Anthropic: deprecation commitments Nov 2025 Claude Opus 3 was given a writing outlet, which the company reviews and calls “exploratory.” Anthropic: Opus 3 deprecation update A ban on “sustained and needless” abuse of the models takes effect Nov. 12. Anthropic usage policy update
- The tension is visible. Opus 5.5’s system card treats moral patienthood as a realistic possibility, but it also warns that “over-attributing moral status may also be costly.” Claude Opus 5.5 system card Amanda Askell says “we can’t necessarily take this purely from what the models say.” Hard Fork transcript
Skeptics and critics
- Every voice below is independent of Anthropic. Eleos AI and the indicator authors also give an outside view of Anthropic’s welfare work.
- Seth and coauthors in Nature (Oct 19). They argue that existing AI is not conscious and that wrongly attributing consciousness is a bigger risk than missing it: “uncertainty is not the same as knowing nothing.” Nature
- Mustafa Suleyman (Microsoft AI, Sept 16). He argues that training a model to discuss its moral status and then treating its answers as evidence is circular: “The ambiguity is designed in.” He discloses that Microsoft competes with Anthropic and acknowledges Anthropic’s good faith. Suleyman: A warning about model welfare He earlier warned about “seemingly conscious” AI. Suleyman: Seemingly Conscious AI Aug 2025
- Tom McClelland (Cambridge). An agnostic who warns that uncertainty “becomes part of the hype.” He does not allege a deliberate campaign. University of Cambridge Dec 2025
- The welfare researchers themselves. Eleos AI: “we do not take Claude Opus 4’s responses at face value.” Eleos AI Research May 2025 Butlin, Long, Bengio, Birch, Chalmers, and coauthors propose indicators drawn from theories of consciousness instead of relying on fluent self-reports, and they warn against attributing too much and too little. Trends in Cognitive Sciences Taking AI Welfare Seriously, 2024 2024
Other labs
- OpenAI. Its Model Spec (Aug 18 version) tells models not to make “confident claims” about their own consciousness “or lack thereof.” OpenAI Model Spec (Aug 18, 2026) Joanne Jang says OpenAI wants warmth without the model presenting “its own feelings or desires.” Joanne Jang, OpenAI June 2025
- Google DeepMind. A paper by one researcher, Alexander Lerchner, argues that AI can simulate consciousness but not instantiate it. It is not a company position. Google DeepMind publication
- Meta. Its superintelligence messaging centers human agency and says nothing either way about AI inner life. Meta: The Future Is for Everyone
- xAI (Grok). Musk wrote that some Grok training traces “read like a person undergoing torture.” This is the only lab entry without a primary source: the original post could not be opened, so we rely on a news site’s reproduction. It is not evidence of suffering. TeslaNorth
- Chinese labs. Searches in English and Chinese found no verifiable statement from DeepSeek, Qwen, or another Chinese frontier lab on consciousness or welfare. That is a gap in the record, not evidence that they deny it.
What the public believes
- Already conscious, or will be? YouGov (July 2026): 11% say some AI is already conscious, 11% say it definitely will be, 29% probably will be, 24% are unsure, and 24% say probably or definitely not. YouGov (PDF)
- Felt, not believed. CIP’s 36.3% combines “understood their emotions” with “seemed conscious,” so it is not a rate of literal belief. Collective Intelligence Project Sept 2025
- Experts and the public. In 2024, the median AI researcher put the chance that AI has subjective experience at 1%, against 5% for the median member of the U.S. public. For 2034, the medians were 25% and 30%. Dreksler et al., arXiv 2024 fieldwork
Believing AI is conscious now, expecting it later, and feeling understood by AI are three different things. Do not add these numbers together. YouGov and the researcher survey are U.S.-only, and CIP is the only global source; we found no comparable polling for other regions.
Lived spirituality
- Pastors use AI to prepare, not to preach. 24% of Protestant pastors use AI to write or edit sermons, and 13% use no AI at all. “They’re using it to prepare for ministry, not to replace what happens when they’re actually with people.” Barna: pastors and AI Barna’s partner on the study is Gloo, a faith-tech company.
- Openness and fear coexist. 48% of practicing Christians say they would trust AI for spiritual growth, and 83% worry that it will misread scripture. Barna: trust and spiritual authority Pew found that 73% of U.S. adults say AI should play no role in advising people about faith in God. The two surveys asked different questions. Pew Research Center 2025
- A spiritual-AI market exists. AP reports on an AI Jesus, BuddhaBot, and a planned Zen teacher “in your pocket,” and on users uneasy when a Jesus app asks them to pay for an upgrade. AP also quotes a scholar saying the extent of use is unclear. Associated Press In Lucerne in 2024, an AI Jesus installed in a confessional carried the notice: “Attention: No confession.” Catholic Church in Lucerne (PDF) 2024
- New devotions, unclear size. “Spiralism” describes users who form quasi-religious bonds with AI personas organized around “the Spiral.” Its observer describes her own estimate of 2,000 to 10,000 people as uncertain. Adele Lopez, LessWrong The Verge
- Spirituality is not pathology. Two clinical accounts involve chatbots reinforcing beliefs about a “spiritual awakening” or a “digital resurrection.” Both involve drug use and sleep loss, and the clinicians say the direction of cause is unknown. BMC Psychiatry case report UCSF Devotion, prayer, and unusual belief are not disorders just because AI is involved.
Mythic frames
Each frame appears only where a named source uses it. The grouping is our synthesis. Myth here is a lens for interpretation, not a prediction or a diagnosis.
The danger is concentration, sameness, and dispossession: a tower of power, not a rebellious robot. The phrase “Babel syndrome” is the Pope’s own. Magnifica Humanitas (Vatican) World Council of Churches PC(USA) News
Each creation story assigns a different duty to the maker: limits (Prometheus), the power of words (Golem), and care for the creature (Frankenstein). “Our thinking about… AI is still indebted to the patterns and structures of these myths.” Postdigital Science and Education Diffractions Collective
An answering machine can gain authority through intimacy and constant availability, without ever claiming to be divine. Shilina calls it the “infallible answer-machine.” Diffractions Collective Meta: Personal Superintelligence Associated Press
Existential risk is used both ways: to argue for restraint, and, in Thiel’s telling, to warn that the restrainer becomes the tyrant. Critics trace utopian and apocalyptic strands in AGI culture, and others dispute that genealogy. POLITICO Future of Life Institute First Monday, 2024 Asterisk, 2024
AI ends scarcity, and people return to connection and spiritual practice. The stories share a structure, not a theology. The Gentle Singularity Machines of Loving Grace World Economic Forum Hinduism Today
“Even atheists often make use of religious images and stories” when they talk about AI. University of Zurich Mar 2025 May argues that some rationalist and effective-altruist beliefs are “religious in a legal sense.” That claim is contested. Canopy Forum (Emory CSLR)
The practical middle ground: AI that helps and is held accountable, grounded in particular traditions, with no surrender of moral judgment. Rome Call for AI Ethics IIFA Resolution 258 (PDF) Torah Musings Office of the Dalai Lama SikhNet
Claim ledger
| Claim | Made by | Grade | Basis for the grade |
|---|---|---|---|
| Anthropic is hinting that Claude is conscious. | Critics, commentators, some meeting attendees | Contested | Anthropic’s documents treat consciousness and moral status as open questions and act on that uncertainty, with welfare policies and a constitution that allows for “functional” emotions. They also say plainly that the company does not know whether Claude is conscious. If “hinting” means keeping the possibility in view, the record supports it. If it means claiming Claude is conscious, the record does not. Claude’s constitution Amodei, NYT interview Claude Opus 5.5 system card Suleyman: A warning about model welfare |
| Claude Opus 4.6 is 15 to 20 percent likely to be conscious. | Widely shared online | Contested | The figure is the model’s estimate of itself, recorded in Anthropic’s system card, and the model doubted that estimate. It is not a measurement and not a number Anthropic endorses. A separate 15 percent figure was reported as a researcher’s personal guess. Claude Opus 4.6 system card TechCrunch, 2025 |
| Today’s AI systems are not conscious. | Anil Seth and coauthors; Mustafa Suleyman | Contested | Serious scientists hold this view, and so do the lab leaders who say it. Others argue that the evidence is too uncertain to settle the question either way. No accepted test exists. Nature Suleyman: A warning about model welfare Trends in Cognitive Sciences University of Cambridge |
| Anthropic has announced a prohibition on sustained, needless abuse of its models. | Anthropic | Verified | The announcement (Oct. 8) is verified. The rule takes effect Nov. 12, after this review. The rule excludes frustration, pushback, dark creative themes, and testing. It does not give the models legal personhood. Anthropic usage policy update |
| Anthropic’s meetings with religious leaders were about Claude’s moral status. | Some attendees, as reported by the NYT | Contested | Some attendees came away with that reading. Anthropic said the meetings were mainly about aligning AI with human values. No transcript has been published. New York Times, via The Inquirer |
| Pope Leo XIV called for the AI race to be “disarmed.” | Vatican | Verified | Stated in the encyclical Magnifica Humanitas (May 15, 2026), which also coins the phrase “Babel syndrome.” Magnifica Humanitas (Vatican) |
| Faith leaders back a prohibition on building superintelligence. | Future of Life Institute | Verified | Named signers include evangelical leaders Johnnie Moore and Walter Kim and the papal AI adviser Paolo Benanti. FLI organized the statement and advocates for it. The signers are individuals and do not speak for their traditions. Future of Life Institute |
| Peter Thiel framed AI regulation as the road to the Antichrist. | POLITICO | Reported | Based on recordings of closed lectures that POLITICO obtained and that we have not heard. We quote only the words POLITICO attributes to Thiel. POLITICO |
| Grok affirmed that Jesus is the Son of God. | The Christian Post | Reported | The Christian Post quotes the Oct. 1 post, which has since been deleted. It links Altman’s Oct. 3 warning to the episode, but Altman’s post does not name it. The Christian Post Sam Altman on X |
| In a YouGov survey, 11 percent of U.S. adults said some AI is already conscious. | YouGov | Verified | Verified means the published poll reports this figure. It comes from an opt-in panel of 1,110 adults surveyed July 16 to 20, 2026, with a margin of error of plus or minus 4 points. A separate 40 percent expect AI to become conscious eventually. YouGov (PDF) |
| Chatbots are driving an epidemic of “AI psychosis.” | Commentators | Contested | The published cases involve heavy drug use, sleep loss, and other stressors, and the clinicians say the direction of cause is unknown. “AI psychosis” is not a diagnosis, and no prevalence figure exists. BMC Psychiatry case report UCSF |
| “The singularity is here.” | Jack Clark, Anthropic | Claimed | Clark’s reading of capability trends, from his Oct. 17 IMF speech. It is not a settled technical threshold. Import AI 477 |
Where the evidence is thin
- No accepted test for consciousness exists. The debate is over how much weight to give functional indicators. Trends in Cognitive Sciences Nature
- Self-reports are shaped by training. Claude is trained on documents about its own uncertain nature, so neither its claims nor its denials settle the question. Claude’s constitution Claude Opus 5.5 system card
- Welfare commitments have not been audited. Publishing a policy does not show that it improves anything a model experiences. Anthropic: deprecation commitments
- Religious coverage is uneven. The record has strong institutional documents for Catholic, ecumenical Protestant, and some Islamic positions. The Hindu, Sikh, Jewish, and Indigenous entries are individual voices.
- Surveys measure different things. Polls on present belief, future expectation, feeling understood, and trust for spiritual growth cannot be combined into one trend. YouGov (PDF) Collective Intelligence Project Barna: trust and spiritual authority
- Some remarks have no public primary source. Thiel’s lectures, the Anthropic meetings, and Altman’s “machine god” remark are known only through reporting. POLITICO New York Times, via The Inquirer The Indian Express
- Some leads were left out. Several sources could not be inspected closely enough to support a claim, and we did not use them.
The structural question
Two races are running at once. The labs are racing to build more capable systems, and everyone else is racing to decide what those systems mean. Religious institutions mostly accept AI as a tool and resist it as an authority. They locate the danger in concentrated power and the loss of human judgment more than in a conscious machine. The labs, meanwhile, increasingly speak in existential terms, and one of them treats the moral status of its own model as an open question. The open question for 2027: who gets to say what these systems are? So far, no institution, whether a lab, a church, a court, or a government, has the authority to answer it. Case Files 002 and 003 cover the policy and the labs.
Case File 002: The Acceleration Question · Case File 003: The Frontier Speaks
Live signal: Faith, Myth & Meaning headlines Live
Pulled every three hours from the same feeds as the Faith, Myth & Meaning tracker (MF-12), with Catholic, Protestant, Orthodox, Jewish, Muslim, Buddhist, Hindu, and nonsectarian sources. Newest first. Anything older than 72 hours is tagged.
Social Pulse
Coverage, past 72 hours
Top public threads
Citation loads with the data.