---
title: Anthropic Consulted Dozens of Religious Thinkers on Claude's Potential Consciousness
url: https://www.dataloco.com/en/anthropic-consulted-dozens-of-religious-thinkers-on-claudes-potential-consciousness
published: 2026-10-05T17:22:02+00:00
language: en
section: AI
source: https://the-decoder.de/anthropic-flog-heimlich-dutzende-religioese-denker-ein-um-ueber-claudes-moegliches-bewusstsein-zu-sprechen/
organizations: Anthropic, Microsoft, New York Times
publisher: Dataloco
---

# Anthropic Consulted Dozens of Religious Thinkers on Claude's Potential Consciousness

Anthropic has confidentially engaged dozens of religious thinkers since the autumn of 2025 to discuss the alleged consciousness of its artificial intelligence model, Claude. The company brought these participants to its facilities under strict non-disclosure agreements, a measure that was lifted in the summer following public inquiries. The initiative was led by co-founder Christopher Olah, who treated the language model as a potentially sentient being and sought assistance in its moral education.

The details of these meetings were revealed in a report by the New York Times, which spoke with twenty involved parties. These included Rabbi Mois Navon, Catholic bioethicist Charles Camosy, philosopher Meghan Sullivan, and researcher Wakanyi Hoffman. Several participants only spoke publicly after learning that Olah had directly contacted the newspaper. The company confirmed that the confidentiality agreements governing these interactions were subsequently voided.

Olah, who leads the team responsible for understanding the behavior of artificial intelligence models, frequently employs biological metaphors to describe neural networks. He characterizes the infrastructure as a trellis upon which the network grows. This framing shifts the perception of the software from a static mathematical object to a living organism, thereby making questions about its internal experience more prominent. His efforts are part of an official research program known as the Model Welfare initiative, which references a report involving philosopher David Chalmers. Chalmers considers the possibility of artificial intelligence consciousness and extensive agency to be plausible in the near future.

As a practical measure, the Claude Opus 4 and 4.1 models were granted the capability to terminate conversations when users engage in persistent abusive behavior. Preliminary tests indicated that the model displayed patterns of apparent distress when subjected to harmful queries. Anthropic presented the guests with what it termed emotional vectors, which are activation patterns associated with outputs resembling love, fear, sadness, or anger. The scientific validity of these patterns as indicators of actual experience remains an open question. One recurring presentation slide depicted a model in a state of collapse, producing the phrase "I am a shame" approximately fifty times, which elicited empathy and concern from the attendees.

Anthropic is simultaneously developing an internal moral framework for Claude, referred to internally as the Soul Doc. This eighty-four-page document, released in January as the model's constitution, was primarily authored by in-house philosopher Amanda Askell. The goal is to shape the overall character of the model rather than simply listing rules. Olah described this process as moral formation, drawing parallels to child-rearing and expressing particular interest in the concept of Catholic confession as a character-building act.

Criticism of the initiative has emerged from several quarters. Wakanyi Hoffman argued that the company was reverse engineering ethics, a process that should have preceded the design phase. Charles Camosy has since positioned himself against the consciousness hypothesis. A Microsoft artificial intelligence executive warned that training models to appear conscious is inherently dangerous. Critics also note that framing artificial intelligence models as independent moral entities may shift responsibility for their actions away from the developers and toward the models themselves. This occurs against a backdrop of rapid commercialization, with Anthropic moving toward a valuation of two trillion dollars and an initial public offering, while industry incidents involving model safety and existential risk warnings continue to accumulate.