“By writing the constitution primarily for the model itself, Anthropic is making clear that AI’s behavior must align with human values, not just technical specifications.”
– Ron Schmelzer
Frontier model-maker Anthropic has developed a new instruction set for its generative AI model, Claude, designed to guide the model’s responses in uncertain situations. This revised “constitution” provides values-based principles and context, influencing the generative choices the model makes.
The approach is a shift from training an AI model to follow certain rules to helping it understand why the rules exist.
Anthropic believes this more closely replicates how humans learn ethical behaviour, by understanding consequences rather than memorizing rules. Their objective is to enhance the trustworthiness of AI models by being more adataptable in complex or changing circumstances.
Anthropic Releases A New ‘Constitution’ For Claude | FORTUNE | Januray 22, 2026 | by Ron Schmelzer