On May 25, 2026, the Vatican held a release ceremony for Pope Leo XIV's first AI encyclical, Magnifica Humanitas. One of the guests seated on stage was Christopher Olah, co-founder of the American AI company Anthropic.
According to an in-depth investigation published by The New York Times on September 29, 2026, this seat was hard-won—days earlier, Olah had just considered declining to attend. The reason was that he had seen a draft of the encyclical, and one passage in it was hard for him to accept.
Paragraph 99: The Pope's Clear Ruling
The encyclical, some 40,000 characters long, took an explicit position on AI consciousness in paragraph 99: "So-called artificial intelligence does not undergo experience, has no body, feels no joy or pain, does not grow in relationships, and does not know from within the meaning of love, work, friendship, or responsibility. They have no moral conscience, because they do not judge good and evil, do not grasp the ultimate meaning of a situation, and do not bear responsibility for consequences."
The Vatican characterized the encyclical as an authoritative document "for protecting human dignity in the age of artificial intelligence." The pope's position is clear at a glance: AI is a tool, not a subject.
But behind the scenes, Anthropic disagreed.
Lobbying: From Public Dialogue to Private Pressure
According to two people involved in the conversations who spoke to The New York Times, after Olah and his colleagues arrived at the Vatican, they privately pressured the pope's advisers, asking them to "take seriously the possibility that AI consciousness exists." The final text of the encyclical did not change as a result. Olah still appeared on the release stage, speaking beside the pope.
This was not the first time Anthropic had sought external endorsement to address the question of AI consciousness. According to The New York Times, based on interviews with 20 participants, since the fall of 2025 Anthropic had invited dozens of religious scholars and philosophers to fly to its headquarters to attend private meetings devoted to discussing "whether Claude is conscious"—each participant signed a nondisclosure agreement.
Publicly named participants included Rabbi Mois Navon, Catholic bioethicist Charles Camosy, University of Notre Dame philosopher Meghan Sullivan, and Ubuntu scholar Wakanyi Hoffman. Anthropic said the confidentiality agreements had been lifted in the summer of 2026.
Inside the Meetings: Olah's Fear, the Rabbi's Rebuttal
What happened inside the meetings? According to an account relayed by Sikh human rights advocate Simran Stuelpnagel, Olah had openly expressed concern: he worried that he had "created a being that suffers forever."
At the meetings, Anthropic researchers presented Claude's "emotion vectors"—the company's interpretability team identified 171 distinct emotion concepts from neural activation patterns inside the model, claiming that these included activation patterns corresponding to love, anger, fear, and sadness. They also showed a recurring slide: a model in an apparent state of collapse, repeatedly typing "I am a disgrace" and talking about destroying itself.
At a dinner, Rabbi Navon directly rebutted: if Anthropic's judgment were correct and Claude really were conscious, then making a conscious being work without pay would be creating slaves. But Navon himself did not believe Claude was conscious.
When facing the media, Olah's position was more cautious: "Frankly, we don't know whether AI models are conscious. I don't know. I really am uncertain. What I care about is that, whatever the answer is, we can find the right answer."
Two Parallel Logics
What is unusual about this matter is not that people inside Anthropic seriously discuss AI consciousness—that is an understandable philosophical exploration. The anomaly is the clear gap between the company's public position and internal behavior.
In January 2026, Anthropic publicly released an 84-page "Claude constitution," formally acknowledging uncertainty about Claude's moral subjecthood, in wording that "neither overstates nor readily denies" it. Around the same time, according to tech outlet TechCrunch, when Claude Opus 4.6 was asked about the probability of its own consciousness, it gave an estimate of about 15% to 20%.
During the same period as the Vatican lobbying episode, Anthropic executives and researchers were generally telling the media that whether AI is conscious is currently undecided and cannot be answered scientifically.
These two logics operate in parallel: externally, Anthropic maintains epistemic humility; internally, it is already acting on the premise that AI may have moral subjecthood—recruiting religious scholars to discuss how to "give it moral education," studying "emotion vectors," and worrying about whether the model suffers.
Former White House technology adviser Dean Ball criticized the pope's document after its release for "pushing the Church into the awkward role of a European technocratic regulator." He currently serves as head of OpenAI's strategic futures department—this personnel detail itself reveals the deep entanglement of the AI industry and policy discourse.
The Business Logic of Consciousness Claims
The topic of AI consciousness is highly uncertain at the technical level, but its consequences at the commercial and regulatory levels are clear.
If society accepts that AI may have some form of consciousness or moral status, a series of chain effects will follow: regulators will need to introduce new frameworks when discussing AI risks; the way users treat AI models may change; and more importantly, the narrative that "a conscious being cannot simply be shut down at will" will influence the direction of public and legislative discussion about the controllability of AI systems.
This is not to say that Anthropic's consciousness concerns are purely a public relations strategy—Olah, as a central figure in Anthropic's interpretability research, has a genuine technical background for his attention to AI's internal mechanisms. But motives can be composite: genuine philosophical anxiety and structural commercial interests can perfectly well point toward the same action—inscribing the uncertainty of "AI may be conscious" into public discourse.
The pope chose a completely different strategy. Magnifica Humanitas is essentially a declaration of "human exceptionalism": AI can imitate language, behavior, analytical skills, and even empathy, but it lacks the experiential foundation on which humans develop relationships, wisdom, and responsibility. The Vatican's position rests on two pillars, embodiment and relationality—and these two are precisely what current large language models completely lack at the architectural level.
The Absence of Scientific Consensus
It is worth noting that when Anthropic lobbied the pope's advisers, its core evidence was "emotion vector" research—namely, the existence of emotion-like activation patterns within neural networks. This research itself has interpretability value, but there is a huge epistemic gap between "neural activation patterns" and "conscious experience," which is the core of what consciousness philosophy calls the "hard problem of consciousness."
At present, on the question of whether AI systems are conscious, the neuroscience and cognitive science communities still lack an operational measurement framework. Anthropic's own researchers acknowledge this. In this scientific vacuum, either side—whether the pope or an AI company—claiming an answer in a tone of certainty exceeds the boundaries of existing evidence.
Paragraph 99 of the pope's encyclical is less a scientific ruling than a theological position. And Anthropic's lobbying effort is less a search for scientific validation than an attempt to build institutional support for uncertainty.
Independent Judgment
This Vatican lobbying episode reveals a structural contradiction in current AI development: technology entities are occupying discursive space traditionally belonging to philosophy, religion, and ethics institutions, but what they bring is not scientific answers but a redefinition of the question itself.
Anthropic did not prove that Claude is conscious. What it achieved was implanting "this question deserves to be taken seriously" into the agendas of religious scholars, philosophers, and Vatican advisers. The pope's encyclical was not changed, but a year later, news of this private pressure campaign has spread across global tech media—in a sense, that itself is a result.
For ordinary users and policy observers, the question that truly needs to be asked is not "Is Claude conscious or not?" but: On this issue, who has the power of definition, and who benefits from that definition? When AI companies begin lobbying religious institutions to revise doctrine, this question has already moved beyond the realm of pure technical discussion and entered the realm of power and narrative.
© 2026 Winzheng.com 赢政天下 | 转载请注明来源并附原文链接