I had a false hit of the "CBYI" rule in a base64 section of a message.
SpamAssassin failed to detect that the section was binary because it said:

Content-Type: MULTIPART/MIXED; BOUNDARY="-559023410-851401618-985897488=:11156"
...
Content-Type: TEXT/PLAIN; charset=US-ASCII
Content-Type: APPLICATION/octet-stream; name="chpost2001.prn"
Content-Transfer-Encoding: BASE64

(note the caps)

According to RFC 1521, this is perfectly legal;
most of these things are case-insensitive.  See
 http://www.freesoft.org/CIE/RFC/1521/4.htm

I note that, even if you change this, I might have the same problem
if someone sent an attachment uuencoded, so:

Other changes that would be good:
  - Ignore uuencoded sections of messages (or expand them before scanning).
  - Change the pattern to /\bCBYI\b/ which still catches nearly all
    (perhaps all) of the CBYI spam I have gotten, but makes it much
    less likely to match a binary.
  - Might also be good to change /HUNZA/ to /\bHUNZA\b/, even though
    five characters is less likely to match than four.