It looks like a previous message with similar content was auto-trained as good, and you didn’t correct SpamSieve by training it as spam (or, more likely, you did correct it, but not until after the second message had been classified).
When SpamSieve thinks that a message is good, it automatically adds the sender to the whitelist. In this case, it though the message contents were interesting, so it also added the message to the corpus. If it turns out that SpamSieve was wrong, and the message was spam, then when you train the message as spam, SpamSieve will undo all of this.
Looking at the log, one problem seems to be that there is a long delay (6 hours in the first case I saw) between when SpamSieve classified a spam message as good and when you corrected it. Due to the auto-training, for those 6 hours SpamSieve was working from incorrect information. This feeds back into more mistakes and more troublesome auto-training. If you cannot correct SpamSieve’s mistakes promptly, you should turn off auto-training.
Another problem is that until a few days ago you were using a version of SpamSieve from last April. Because of this and because SpamSieve’s corpus is rather old and large, it would probably help to reset the corpus and re-train SpamSieve with a smaller number of recent messages.