The problem is that it's easy to upload 1000 randomly generated phone numbers and get back results. But I think you can distinguish random or sequentially gener

The problem is that it's easy to upload 1000 randomly generated phone numbers and get back results. But I think you can distinguish random or sequentially generated numbers from a real user's actual contact list, by looking at the distribution of friends among the contacts. A real contact list should have a high density of friendships between contacts. The system should only return results if the uploader provides a subgraph that's more highly connected than a random one would be. You'd need to look at real data and tune the parameters to make this effective.