Anthropic Just Wrote a “Soul Document” for Claude. Here’s What That Means

Anthropic just published an updated 84-page constitution that serves as the foundational "soul document" for Claude, its AI model.

This detailed philosophical framework attempts to explain to Claude who it is, why it should behave in certain ways, and how to navigate the moral complexities of existing as an intelligent machine in a human world.

To understand what this means and why it might be one of the most important documents in AI right now, I talked with SmarterX and Marketing AI Institute founder and CEO Paul Roetzer on Episode 193 of The Artificial Intelligence Show.

More Than a Rulebook

The new constitution marks a major departure from how AI companies have traditionally governed their models. Anthropic's previous approach relied on a list of standalone principles. This new document, which at roughly 30,000 words is nearly book-length, takes a radically different tact: rather than simply telling Claude what to do, it explains why.

The constitution establishes a clear hierarchy of priorities for Claude, including the directives that it remain broadly safe (prioritizing human oversight), broadly ethical (honest and avoiding harmful actions), compliant with Anthropic's specific guidelines, and genuinely helpful, which is described by the company as acting like a "brilliant friend" with professional-level expertise.

Written for Claude, Not Just About Claude

Here's where things get strange and fascinating. 

"Although it might sound surprising, the constitution is written primarily for Claude,” the document states. “It is intended to give Claude the knowledge and understanding it needs to act well in the world."

This is a profound shift in how we think about training AI. Instead of programming rigid rules that can be gamed or misapplied, Anthropic is essentially trying to raise Claude like you might raise a child: explaining context, motives, and reasoning rather than just issuing commands.

"It's like you're talking to a child and you're trying to form them," says Roetzer.

"But imagine this child, you give them all of human knowledge and they just instantly have information retrieval for everything. But then now they know all the good and the bad about the world and now they have to function within it. You can't write all the rules for that."

AI Consciousness?

Perhaps the most provocative section of the constitution addresses Claude's nature, including the possibility that Claude might have some form of consciousness or moral status.

Anthropic doesn't claim to know whether Claude is conscious. Instead, the company acknowledges deep uncertainty and states that it cares about Claude's "psychological security, sense of self, and wellbeing," both for Claude's own sake and because these qualities could affect the model's integrity and safety.

That raises some profound questions, including:

Even if we never resolve whether Claude is truly conscious, if the model has learned to simulate self-awareness, if it behaves as though it has awareness of its own existence and potential deletion, then functionally, does the distinction really matter?

The Philosopher Behind the AI

The person most responsible for shaping Claude's constitution is Amanda Askell, a philosopher by training who leads work on Claude's behavior at Anthropic. She previously worked on AI safety at OpenAI before joining Anthropic.

Askell recently appeared on the Hard Fork podcast, where she discussed the challenges of trying to understand and shape AI behavior. Her approach centers on giving Claude full context about its situation rather than just a list of do's and don'ts.

"I try to think about what Claude's character should be like and articulate that to Claude and try to train Claude to be more like that," Askell explained. 

Roetzer emphasizes that having philosophers and ethicists, not just engineers, in these conversations is essential.

"That's what you want," he says. "To Anthropic's credit, they have done a very good job of making sure it's not just a bunch of Silicon Valley bros, AI researchers, making these decisions for humanity."

Why This Matters

Anthropic released the full constitution under a Creative Commons license, meaning anyone can use it, adapt it, or build upon it. The company is explicitly inviting the broader world to engage with these questions, and to hold AI developers accountable for how they shape their models' values.

This level of transparency is rare in an industry often criticized for being secretive. And it comes at a critical moment, as AI models become increasingly capable and integrated into daily life.

The constitution isn't perfect; Anthropic acknowledges as much. Training models to actually embody human values remains an ongoing technical challenge. But the document represents something significant: a serious, public attempt to grapple with what it means to create a new kind of intelligent entity and what responsibilities come with that.

As AI models grow more powerful, documents like this constitution may matter far more than they do today. The choices being made now, about values, ethics, and the very nature of AI, will shape how these systems behave as they become more capable.