Claude 4.5's constitution expands to 200+ principles with automated refinement
Anthropic disclosed that Claude 4.5 was trained against a written constitution containing over 200 principles, up from ~50 in the original Constitutional AI paper. Automated refinement processes update the constitution in response to observed failure modes.
The 200-principle constitution is structured hierarchically — broad principles like "be honest" decompose into 8-15 more specific operational principles each. New principles can be added in response to identified failure modes without retraining from scratch.
The headline outcome: a 40% reduction in alignment failures versus the static 50-principle constitution, measured on Anthropic's internal red-team suites. The automated refinement is what makes the approach scale — manual review of 200 principles per release cycle would be intractable.