AI will need a social contract: clear rights for users, responsibilities for builders, and meaningful recourse when systems cause harm.