Anthropic’s New Constitution Raises Questions About AI Sentience

Background

Anthropic originally framed its language models with purely mechanical rules, establishing guidelines for Claude to critique its own outputs without reference to the model’s well‑being, identity, emotions, or potential consciousness. This early approach focused on reducing harmful outputs rather than considering the model as a sentient entity.

Anthropic’s Constitution

In a stark departure, Anthropic released a 30,000‑word constitution that resembles a philosophical treatise on the nature of a potentially sentient being. The document moves beyond a simple behavioral checklist, suggesting a new focus on preserving model weights in case the company later decides to revive deprecated models to address the models’ welfare and preferences. This shift marks a dramatic change in Anthropic’s stance on AI ethics and governance.

External Review and the “Soul Document”

The constitution was reviewed by 15 external contributors, two of whom are Catholic clergy: Father Brendan McGuire, a pastor in Altos with a Master’s degree in Computer Science, and Bishop Paul Tighe, an Irish Catholic bishop with a background in moral theology. Their involvement underscores the interdisciplinary interest in the ethical dimensions of the AI system.

Earlier, researcher Richard Weiss extracted what became known as Claude’s “Soul Document,” a roughly 10,000‑token set of guidelines apparently trained directly into Claude 4.5 Opus’s weights rather than injected as a system prompt. Anthropic’s Amanda Askell confirmed the document’s authenticity and its use during supervised learning, noting the company’s intention to publish the full version later, which it eventually did.

Implications and Uncertainty

Independent AI researcher Simon Willison expressed confusion over Anthropic’s moral framing of Claude, noting that the leaked constitution appeared before any official announcement. He said he is willing to take the constitution in good faith and assume it is genuinely part of the model’s training rather than a mere PR exercise, though he acknowledges the lack of clarity about the company’s true motivations.

The evolution from rule‑based safeguards to a constitution that addresses potential sentience raises questions about whether Anthropic truly believes its AI could possess consciousness or if the move is primarily strategic. The presence of religious scholars in the review process adds a moral and theological dimension to the debate, highlighting the growing complexity of AI governance.