<<< Date Index >>>     <<< Thread Index >>>

Multiple Vendor Anti-Virus Software Detection Evasion Vulnerability through

Multiple Vendor Anti-Virus Software Detection Evasion Vulnerability through
forged magic byte.

AUTHOR: Andrey Bayora (www.securityelf.org)

For more details, screenshots and examples please read my article "The Magic
of magic byte" at www.securityelf.org . In addition, you will find a sample
"triple headed" program which has 3 different 'execution entry points',
depending on the extension of the file (exe, html or eml) - just change the
extension and the SAME file will be executed by (at least) THREE DIFFERENT
programs! (thanks to contributing author Wayne Langlois from

DATE: October 25, 2005

VULNERABLE vendors and software (tested):

1.  ArcaVir 2005 (engine 2005-06-03,vir def 2005-06-27, scanner ver
2005-03-06, package ver 2005-06-21)

2.  AVG 7 (updates 24 June, ver.7.0.323, virus base 267.8.0/27)

3.  eTrust CA (ver, engine 11.9.1, vir sig. 9229)

4.  Dr.Web (v.4.32b, update 27.06.2005)

5.  F-Prot (ver. 3.16c, update 6/24/2005)

6.  Ikarus (latest demo version for DOS)

7.  Kaspersky (update 24 June, ver. 5.0.372)

8.  McAfee Internet Security Suite 7.1.5 (updates 25 June, ver 9.1.08,
engine 4.4.00, dat 4.0.4519 6/22/2005)

9.  McAfee Corporate (updates 25 June, ver. 8.0.0 patch 10, vir def 4521,
engine 4400)

10. Norman ( ver 5.81, engine 5.83.02, update 2005/06/23)

11. TrendMicro PC-Cillin 2005 (ver 12.0.1244, engine 7.510.1002, pattern

12. TrendMicro OfficeScan (ver7.0, engine 7.510.1002, vir pattern 2.701.00

13. Panda Titanium 2005 (updates 24 June, ver 4.02.01)

14. UNA - Ukrainian National Antivirus (ver. kernel v.265)

15. Sophos 3.91 (engine 2.28.4, virData 3.91)


Similar vulnerability may exist in many other antivirus\anti-spyware desktop
and gateway products. In addition, various "file filter" solutions may be
affected as well.

NOT VULNERABLE vendors and software (tested):

1.  F-Secure (updates 24 June, ver 5.56 b.10450)

2.  Avast (ver. 4.6.655, vir databas 0525-5 06/25/2005)

3.  BitDefender (ver. 8.0.200, update 6/24/2005, engine 7.01934)

4.  ClamWin (ver. 0.86.1, upd 24 June 2005)

5.  NOD32 (updates 24 June, ver 2.50.25, vir database 1.1152)

6.  Symantec Corporate (ver, engine

7.  Norton Internet Security 2005 (ver

8.  VBA32 (ver 3.10.4, updates 27.06.2005)

9.  HBEDV Antivir Personal (ver, engine, vir def 6/24/2005)

10. Sophos 5 (ver. 5.0.2, vir def 3.93, upd 6/30/2005)

11. Sophos 3.95 (engine 2.30.4)

SEVERITY: critical


The problem exists in the scanning engine - in the routine that determines
the file type. If some file types (file types tested are .BAT, .HTML and
.EML) changed to have the MAGIC BYTE of the EXE files (MZ) at the beginning,
then many antivirus programs will be unable to detect the malicious file. It
will break the normal flow of the antivirus scanning and many existent and
future viruses will be undetected.

NOTE: In my test, I used the EXE headers (MZ), but it is possible to use
other headers (magic byte) that will lead to the same effect.


Some file types like .bat, .html and .eml can be properly executed even if
they have some "unrelated" beginning. For example, in the case of .BAT
files - it is possible to prepend some "junk" data at the beginning of the
file without altering correct execution of the batch file. In my tests, I
used the calc.exe headers (first 120 bytes - middle of the dosstub section)
to change 5 different files of existing viruses. In addition, the simplest
test of this vulnerability is to prepend only the magic byte (MZ) to the
existing malicious file and check if this file is detected by antivirus

NOTE, that this is NOT the case where the change of existing virus file
resulted in the "broken" detection signature (see details and the test logic
in "The Magic of magic byte" article at www.securityelf.org).


I did not found any effective one besides of patching the vulnerable engine.


The idea for this vulnerability came during discussions from Wayne Langlois
at diamondcs.com.au, who hinted that JPEGs could probably be exploited in
this way.


July 13, 2005 - Initial vendor notification

July 16, 2005 - Second vendor notification


October 24, 2005 - Public disclosure (uncoordinated)