Anthropic updates Claude’s constitution with new safety guidelines

On Wednesday, Anthropic released a revised version of Claude’s Constitution, an 80-page document that provides a holistic explanation of the context in which the Claude chatbot operates. The release of the document, which is structured into four separate parts, was timed with Anthropic CEO Dario Amodei’s appearance at the World Economic Forum. This update reinforces the company’s reliance on Constitutional AI. Anthropic, which was co-founded by Jared Kaplan, defines Constitutional AI as a training method that uses ethical principles rather than human feedback to guide the model’s behavior.

The updated document retains most of the same principles first published in 2023, nearly three years ago, after an initial policy memo in 2022. However, the revised version adds more nuance and detail regarding ethics and user safety. According to Anthropic, these principles guide the model to take on normative behavior and avoid toxic or discriminatory outputs. Under these safety guidelines, discussions of developing a bioweapon are strictly prohibited. Additionally, the document instructs the model to direct users to relevant emergency services or offer basic safety information during situations that involve a risk to human life, even if it cannot provide further details.

The document also outlines how these principles theoretically impact Claude’s behavior in real-world scenarios. In the section on ethical practice, the document notes that the company is less focused on the chatbot’s ability to theorize about ethics and more concerned with its practical ability to act ethically in specific real-world contexts. Beyond practical safety, the document addresses the philosophical nature of artificial intelligence, stating that “Claude’s moral status is deeply uncertain,” according to Claude’s Constitution. Through this framing, Anthropic positions itself as the ethical alternative to competitors like OpenAI and xAI, though some might argue this positioning is intended to portray the company as a more restrained business.

Why it matters

Anthropic is using this update to solidify its brand as the restrained, ethical alternative to competitors, while signaling its engagement with high-level philosophical questions about AI consciousness.