As I understood it they were using machine learning models to identify attacks. Abuse and fraud strategies are constantly evolving, so the benefit of having ma

As I understood it they were using machine learning models to identify attacks. Abuse and fraud strategies are constantly evolving, so the benefit of having many clients is that once you see a new attack strategy at one, you can instantly protect all your clients from that threat. In your own example "Twitter likely doesn't want to allow people to be able to test against their spam shields by spamming other people" Yes, the attacker may find a new pattern that can bypass the shield, but as soon as that pattern is added to the machine learning recognition, they are then protected against it on their own system. "spam detection is just pattern recognition" The more patterns you can analyse, the better your recognition?