Auto Train causing issues and increased SPAM??

Can anyone tell me why the following is happening?

SS seem to be predicting SPAM as good as well as Auto Training SPAM as good. There must be an easy solution to this. When I first started using this application for the first 3 weeks it worked well but it has been an uphill battle since then.

Thanks,
Andrew.

------------examples from logfile----------------------------
Predicted: Good (27)
Subject: Great opportunity to give her a real pleasure
From: dwursm@urs.us
Identifier: yIENBrLXBQRfD5I6me6rGA==
Reason: P(spam)=0.500[0.500], bias=0.953,
Date: 2008-05-19 08:07:59 -0400

Trained: Good (Auto)
Subject: Great opportunity to give her a real pleasure
Identifier: yIENBrLXBQRfD5I6me6rGA==
Actions: added rule <From (address) Is Equal to “dratcliffe@cogeco.ca”> to SpamSieve whitelist, added rule <From (address) Is Equal to "dwursm@urs.us"> to SpamSieve whitelist, added rule <From (name) Is Equal to “Harvey Zavala”> to SpamSieve whitelist
Date: 2008-05-19 08:07:59 -0400

Launched: 2.6.6
Date: 2008-05-19 14:50:20 -0400

Predicted: Good (27)
Subject: Exquisite luxury footwear and designer bags with 80% discount
From: qtdyrtriqrkm@blumenthal.com
Identifier: p7edGiQUEh5eVqBPRGpQCA==
Reason: P(spam)=0.500[0.500], bias=0.953,
Date: 2008-05-19 14:50:36 -0400

Trained: Good (Auto)
Subject: Exquisite luxury footwear and designer bags with 80% discount
Identifier: p7edGiQUEh5eVqBPRGpQCA==
Actions: added rule <From (address) Is Equal to "qtdyrtriqrkm@blumenthal.com"> to SpamSieve whitelist, added rule <From (name) Is Equal to “Heriberto Jimenez”> to SpamSieve whitelist
Date: 2008-05-19 14:50:37 -0400

Predicted: Good (27)
Subject: Change your sex life with VPXL!
From: redevelopsl6@cnc-speedform.de
Identifier: S2KrviGvpTnmOUsziRQZKQ==
Reason: P(spam)=0.500[0.500], bias=0.953,
Date: 2008-05-19 14:50:37 -0400

Trained: Good (Auto)
Subject: Change your sex life with VPXL!
Identifier: S2KrviGvpTnmOUsziRQZKQ==
Actions: added rule <From (address) Is Equal to "redevelopsl6@cnc-speedform.de"> to SpamSieve whitelist, added rule <From (name) Is Equal to “Carey Mccarthy”> to SpamSieve whitelist
Date: 2008-05-19 14:50:37 -0400

If SpamSieve thinks that a message is good, it’s normal for it to try to learn from that message (e.g. by adding the sender address to the whitelist). It would be a lot of work if you had to specifically tell SpamSieve that the message was good, so it’s usually better if SpamSieve just assumes that it was right. That’s what auto-training is all about. If it turns out that SpamSieve was wrong, then you would correct it by training the message as spam, and all would be good again (e.g. it will disable the whitelist rule).

One potential problem is if you don’t correct SpamSieve promptly. In that case, there would be a long window of time in which SpamSieve had (incorrectly) trained itself that a spam message was good, so it would be making predictions based on false information. If you cannot correct SpamSieve promptly, you should turn off auto-training.

Have you been training the spam messages that get through as spam? If so, there should be “Trained: Spam (Manual)” entries for them in the log.

How many good and spam messages are in the corpus, as shown by the Statistics window?