Michel Justen at Asterisk Magazine:
Claude’s constitution is Anthropic’s “final authority” on how Claude should behave. Released in January 2026, the (roughly) 30,000-word document describes how Claude should be broadly safe, broadly ethical, and genuinely helpful. But unlike OpenAI’s analogous Model Spec, which details specific rules that ChatGPT should obey, Claude’s constitution details a character.
Anthropic’s Amanda Askell, lead author of the constitution, thinks of this ideal character as “the well-liked traveler” — the type of virtuous, non-threatening person who can navigate many different cultures and contexts while staying true to his or her values. As Amodei writes: “Claude’s constitution presents a vision for what a robustly good Claude is like; the rest of our training process aims to reinforce the message that Claude lives up to this vision.”
Yet for all of the fanfare around the constitution and the underlying bet on AI “character training,” researchers argue the public knows surprisingly little about how Anthropic trains Claude to follow the constitution — and how well it works.
More here.