For the full details, please see the discussion on Meta Stack Exchange: Can a machine be taught to flag spam automatically?
In short, a group of users have created an application that, historically, has much higher accuracy in identifying spam. This application has been turned on for all sites in the network.
Please direct any general discussion of this new application to the above Meta Stack Exchange discussion. Any discussion or questions as it pertains to Software Engineering is appropriate here - if it needs an answer, I can ping the right people.