Breaking

Anthropic Reportedly Used NDA-Bound Religious Consultations to Debate Claude's Values

October 1, 2026 at 15:33 EDT

Foundation Models
Policy & Regulation

Anthropic reportedly convened confidential meetings with religious and philosophical thinkers around the world as it explored how Claude should express morals and values, and how seriously to take the possibility of AI consciousness. The talks put a frontier model's character work into unusually public view, raising a sharper question for the AI industry: who gets to define a chatbot's values before it mediates sensitive conversations at scale?

Sept. 30, 2026 · Anthropic

Claude's Moral Formation Goes Public

Private consultations with religious and philosophical thinkers have pushed Anthropic's Claude alignment work into a debate over values, consciousness and accountability.

pages in Claude's reported constitution

700K

anonymized Claude.ai conversations analyzed

3,307

AI values identified in research

>15

religious and cross-cultural groups

Scale Check: Values Measured In The Wild

308K

subjective conversations

700K

total sample

Each block is about 100,000 conversations. Real-world data helps reveal deployed behavior, but it cannot certify a model before release.

How The Work Flows

Constitution

Claude's written value framework

→

Wisdom Circles

Outside moral traditions

→

Model Behavior

Pushback, advice and self-description

Why Supporters Care

Open-ended moral cases cannot all be solved with fixed rules, so outside traditions may add hard-won reasoning about character, grief and harm.

Why Critics Worry

Private meetings and consciousness language can blur accountability when the public still cannot see how advice changes a frontier model.

The Unanswered Question

Anthropic says the conversations continue, but has not detailed how the input has changed Claude's training. The live issue is not whether Claude expresses values, but how those values are chosen, tested and challenged.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.