On 9 Apr 2002 at 10:07, Steve Schear wrote: > New breed spam filter slashes junk email > 10:31 09 April 02 NewScientist.com news service > > A new breed of spam-filtering technology that combines peer-to-peer > communications with machine learning could intercept nearly all unwanted > email, according to its creators. > > http://www.newscientist.com/news/news.jsp?id=ns99992141 > > Sounds like it should work quite well at eliminating spam targeted directly at the user. Probably not much risk of an actual personal message looking enough like s spam message to get flagged.
But for distribution lists I think there's substantial risk. Potentially would-be censors could block posts as alleged spam. Also, there's a major security concern. The article didn't say whether users would have to keep a complete copy of the spam database on their local machines or whether they'd have to upload each mail message to the servers with the database, but I think they'd have to do one or the other, and each has obvious drawbacks. (It should be safe to just upload a hash of each message received and compare that to the database, but even that has some risks, and besdies, I got the impression they wanted to do a more thorough comparison. Checking hashes could easily be defeated by appending a separate random string to each copy of the message anyway). All in all, I vastly prefer hashcash. George