Anthropic drafts a ‘constitution’ for Claude

ANTHROPIC HAS PUBLISHED a new set of behavioural guidelines for its Claude AI models covering ethics and safety while also laying out guidance for a scenario in which an Al system becomes sentient.

- KARAN MAHADIK

The newly drafted ‘constitution for Claude’ serves as a final blueprint for shaping the AI models into what the company wants them to be. It provides a description of Anthropic’s vision for the models’ values and behaviour.

Anthropic plans on using the abstract ideals defined in the constitution at various stages of the train-