Anthropic has updated Claude's "constitution," shifting from strict restrictions to broad concepts.

Anthropic has updated Claude's "constitution," moving from strict restrictions to broad concepts

Anthropic has introduced an updated version of Claude's constitution. The company explains that while strict prescriptions ensure predictable behavior, they simultaneously limit the AI's potential. The new concept bets on the model's ability to exercise sound judgment, generalize from experience, and avoid stereotypical responses. According to the developers, it's important not just to force the AI to mechanically execute commands, but to help it understand the motives behind human expectations—this is what will ensure adequate behavior in the real world.

The foundation of the updated constitution consists of four generalized principles: broad safety, broad ethics, alignment with Anthropic's guidelines, and true helpfulness. Ethics, for example, is interpreted as honesty, adherence to good values, and avoiding inappropriate, dangerous, or harmful actions.

Although the public version of the document appears quite general, the company emphasizes that a significant portion of the text is dedicated to detailed explanations for each principle.

A section dedicated to the nature of Claude itself deserves special attention. Anthropic included it in the document due to the uncertainty surrounding the question of possible AI consciousness or moral status—both present and future. The company hopes that clearly defining these aspects in the foundational document will help protect the chatbot's "psychological safety, self-perception, and well-being." This wording highlights the developers' desire to proactively define ethical boundaries for interacting with advanced AI systems.

The announcement of the update came a day after Dario Amodei, founder and CEO of Anthropic, spoke at the World Economic Forum panel discussion "The Day After AGI." In his presentation, he suggested that by 2027, artificial intelligence could reach the level of Nobel laureates in many fields.

Anthropic emphasizes that disclosing details about Claude's operation is happening according to the company's internal schedule. Publishing the full version of the constitution was planned from the outset and is part of its transparency strategy.