Mail with metaphorical language to pubs/coffee shops/high traffic businesses which are hubs for local communities of the minority population. At the hubs, word of mouth could be used.
In 1998, the Internet was quite new, not available to many people aside from university students and primitive dial up access, and not easily available globally. Computer ownership and usage was far from universal in 1998 in many parts of the world. Communities might have started to think about online options, but wouldn't be ready to adopt them yet. Also, in 1998, communities would still be aware aware about the risks of major failures of security in electronic communications like ENIGMA during World War II, and would be wary of it.
Equally important, these populations would not be early adopters of new technologies. They would have been around for centuries, at least, and once they found something that worked, wouldn't be quick to change it.
But, mail was widely available everywhere in the world to even regional small towns in the most remote areas, in 1998 and had been for a long time.
It was secure enough that only a thin veil of metaphor rather than a hard core code (or perhaps a language specific to the minority, a bit like expatriate communities communicating via their homeland language) would be sufficient to avoid casual penetration of the network. Another way to cloak the messages without fully and elaborately coding them would be to use religious metaphors, so that any attempt to penetrate the network could be attacked as religious persecution.
Mail can be sent in untraceable ways with, for example, false return addresses and street corner mail boxes. It isn't expensive even worldwide and can get to even remote places like Papua New Guinea in a week or two. And, mail is a very old technology. Early versions of mail-like information sharing networks existing in the Bronze Age (we have cuniform letter exchanges in the archaeological record). It would be a lot less costly than personally sending word of mouth messages on a regular basis across the world. Messages could be destroyed once received.
There could be a smaller subnetwork of hubs to which news was reported, with with a cell-like structure, with two or three core hubs per territory that would know the addresses of the local hubs in their territory and one core hub per territory. So, even a catastrophe would not take out the entire network, just disrupt one territory's communications somewhat.
Most of the eight million members would know only one or two local hubs and some lore about how to find new hubs if their's were compromised somehow. Perhaps, 80,000 members would know one set of core hub addresses and how to understand the metaphors, perhaps 8,000 or fewer people would know all the hub addresses in a territory and up to about half of the global core addresses. No one outside a territory would know any significant share of hub addresses in other territories, and no one (or at least not more than a dozen or so people) would know all of the core addresses outside their territory.
Also, if you limited your mailing list to thousands of pubs or restaurants/coffee shops (in places, e.g., in the Islamic world where alcohol isn't served) that served as community hubs, then you also aren't revealing the locations of any of the members of the community if the hub is compromised for some reason. There would be a policy of not keeping lists of members, although sometimes the hub would have, for example, a billing address for a member who was a vendor or customer that was buried with other random third-party vendor or customer information. But the list of members served by the hub would only be in the operator's memories.
Yet, this system could spread news worldwide within a month or two to the most distant corners of the world from anyone who could get net to a hub, and in places with faster mail service, or for news only relevant to a single territory and not spread worldwide, it could receive news and distribute it all over the territory or faster news area within a week or so.
Coded messages in mass media (like personal ads or widely distributed artwork) would be less secure. Someone could break the code, and once someone cracked the code, the entire network would be compromised and would have to be rebuilt from scratch, not just discontinuing mailings to one or two hubs out of thousands that could be quickly cutoff.
A variety of means could be used for insiders to identify hubs, from subtle symbols (perhaps keyed to senses that the minority has more acutely than normal humans) to continuous word of mouth updating from one member of the community to another.
Fictional and Real World Examples
The notion of a coffee shop hub network for most members of the community has been explored, for instance in the anime/manga Tokyo Ghoul and to a lesser extent with pubs in the Harry Potter mythos.
A variant of this, used in Kate Elliott's Cold Fire series, would be for correspondence from core hubs to go to (and some core hubs themselves to be) law firms that would then disseminate the news to community hubs like pubs and coffee shops orally. The large volume of correspondence going in and out of law firms, and the discretion and confidentiality of law firms, would be useful in this kind of network.
Many networks for distribution of controlled substances also work this way, with the front business being a drop point for drug pickups that are hidden by the legitimate business that many people who have no knowledge of the covert activities also use.
Using an insider in a public business as a way of communicating secret messages is also explored in Spy x Family (with deeper use of codes), but the larger the community the less likely it is that a common code will be useful. Since 20th century mail was even more secure than it is now, so tough codes for seemingly innocent letters to hubs that get lots of mail in the course of business from all sorts of third-parties isn't necessary.