ANTIVIRUS INTELLIGENCE TO COLLECTIVE - PANDA'S TECHNOLOGY EVOLUTION

Page created by Monica Gallagher
 
CONTINUE READING
ANTIVIRUS INTELLIGENCE TO COLLECTIVE - PANDA'S TECHNOLOGY EVOLUTION
FROM TRADITIONAL
                        ANTIVIRUS
                        TO COLLECTIVE
                        INTELLIGENCE
                        PANDA'S TECHNOLOGY EVOLUTION

                        Technology Paper by Panda Research
                        research.pandasecurity.com

www.pandasecurity.com
FROM TRADITIONAL ANTIVIRUS TO COLLECTIVE INTELLIGENCE    Pag. 2

Content

1 Abstract                                                  3

2 The malware landscape                                     4
2.1 Antivirus laboratories under attack                     4
2.2 Malware techniques and design                           5
    2.2.1 Targeted Attacks: staying below the radar         5
    2.2.2 Malware QA                                        5
    2.2.3 Rootkits and sandbox detection techniques         6
    2.2.4 Runtime-packers                                   6
    2.2.5 Botnets                                           7
    2.2.6 Staged infection vectors                          8
    2.2.7 "Malware 2.0"                                     8

3 Panda’s Technology Evolution                              9
3.1 First Generation: Antivirus                             9
3.2 Second Generation: Anti-malware                         9
3.3 Third Generation: Proactive technologies               10
    3.3.1 Uncloaking techniques                            10
    3.3.2 TruPrevent® Behavior Analysis                    10
    3.3.3 TruPrevent® Behavior Blocking                    12
    3.3.4 Genetic Heuristics                               13
3.4 Collective Intelligence                                14
    3.4.1 Benefiting from Community Knowledge              14
    3.4.2 Automated Malware Protection Process             15
           3.4.2.1 Automated malware collection            15
           3.4.2.2 Automated malware classification        15
           3.4.2.3 Automated malware remediation           16
    3.4.3 Gaining Knowledge on Malware Techniques          16
    3.4.4 Deploying Security Services “from-the-cloud”     17
    3.4.5 A note on white-listing                          17

4 Conclusion                                              19

5 References                                              20
FROM TRADITIONAL ANTIVIRUS TO COLLECTIVE INTELLIGENCE                                     Pag. 3

1. Abstract

T
        here is more malware than ever being released in the wild, and antivirus companies
        relying on signatures to protect users cannot keep up with the pace of creating
        signatures fast enough. As a result, the current installed base of anti-malware solutions
is proving to be much less effective against the vast amounts of threats in circulation.

As we have been able to prove in a recent research study , even users protected with anti-
malware and security solutions with the latest signature database are infected by active
malware. Complementary approaches and technologies must be developed and
implemented in order to raise the effectiveness to adequate levels.

This paper presents the fourth generation of security technologies by Panda Security, called
Collective Intelligence. The Collective Intelligence allows us to maximize our malware
detection capacity while at the same time minimizing the resource and bandwidth
consumption of protected systems.

The Collective Intelligence represents an approach to security radically different to the
current models. This approach is based on an exhaustive remote, centralized and real-time
knowledge about malware and non-malicious applications maintained through the
automatic processing of all elements scanned.

One of the benefits of this approach is the automation of the entire malware detection
and protection cycle (collection, analysis, classification and remediation). However
automation in and by itself is not enough to tackle the malware cat-and-mouse game.
With large volumes of malware also comes targeted attacks and response time in these
scenarios cannot be handled by automation of signature files alone.

The other main benefit that the Collective Intelligence provides is that it allows us to gain
visibility and knowledge into the processes running on all the computers scanned by it.
This visibility of the community, in addition to automation, is what allows us to tackle not
only the large volumes of new malware but also targeted attacks.

                                                                     Written and reviewed by:
                                                  Pedro Bustamante, Senior Research Advisor
                                                         Iñaki Urzay, Chief Technology Officer
                                                   Luis Corrons, Technical Director PandaLabs
                                             Josu Franco, Director of Corporate Development
FROM TRADITIONAL ANTIVIRUS TO COLLECTIVE INTELLIGENCE                                                     Pag. 4

2. The malware landscape

I
    t is a known fact by all security professio­     analysts at the labs”3 or by advocating for
    nals that there are more malware sam­            stronger intervention4 by Law Enforcement5
    ples infecting users than ever before.           to help ease the workload by convicting
                                                     the most active malware creators.
Malware writers have realized they can gain
large amounts of money from distributing                           Unique Samples Received at PandaLabs

malware. The shift in motivation for creating         14.000

malware, combined with the use of                     12.000
advanced techniques, has resulted in an
exponential growth of criminally                      10.000

professional malware being created and
distributed to infect unsuspecting users.
                                                       8.000

                                                       6.000

Also known as a type of targeted attacks,
this new malware dynamic has become the                4.000

next big plague for users and companies                2.000
alike. Gartner estimates that by the end of
2007 75% of enterprises will be infected                  0
                                                               April 2004
                                                               May 2004
                                                               June 2004
                                                                July 2004
                                                               Aug 2004
                                                               Sept 2004
                                                                Oct 2004
                                                                Nov 2004
                                                                Dec 2004
                                                                 Jan 2005
                                                                Feb 2005
                                                                Mar 2005
                                                               April 2005
                                                               May 2005
                                                               June 2005

                                                               Sept 2005
                                                                July 2005
                                                               Aug 2005

                                                                Oct 2005
                                                                Nov 2005
                                                                Dec 2005
                                                                 Jan 2006
                                                                Feb 2006
                                                                Mar 2006
                                                               April 2006
                                                               May 2006
                                                               June 2006
                                                                July 2006
                                                               Aug 2006
                                                               Sept 2006
                                                                Oct 2006
                                                                Nov 2006
                                                                Dec 2006
                                                                 Jan 2007
                                                                Feb 2007
                                                                Mar 2007
with undetected, financially motivated,
targeted malware that evaded their
traditional perimeter and host defenses 2 .          Figure 1:
                                                     Unique samples received at PandaLabs 2004 to 2007

                                                     Initiatives to get law enforcement more
2.1 Antivirus laboratories                           involved are definitely a necessary step in
under attack                                         the right direction but unfortunately it
                                                     seem as an insufficient solution for the
Nowadays antivirus laboratories are under a          short term as the number of variants is
constant and increasingly frequent distributed       increasing incrementally and most of the
denial of service attack. The security industry is   time only the “mules” and “script kiddies”
literally being saturated with thousands of          are actually convicted.
new malware samples every day.
                                                     The more advanced malware writers, who
Each one of these new samples needs to be            are selling their code to spammers, mafias
looked at by an analyst trained in reverse           and criminals, are more evasive and harder
engineering in order to create a signature,          to catch. In addition, the lack of resources
which is costly and resource intensive from a        at most law enforcement agencies around
corporate and business perspective.                  the world, tied to insufficient international
                                                     cooperation and coordination among them
Some companies are trying to deal with the           make for a difficult task when trying to
problem by increasing the number of                  arrest a suspect or known cyber criminal. In
FROM TRADITIONAL ANTIVIRUS TO COLLECTIVE INTELLIGENCE                                      Pag. 5

the long run both a technological and a             In the past a single virus or worm was res­
social approach are needed if we want to            ponsible for infecting hundreds of thou­
solve this problem.                                 sands and even millions of computers. Visi­
                                                    bility of these situations was very obvious
In addition, malware writers are getting            for antivirus labs.
sophisticated and reverse engineering some of
the latest common threats requires a higher         Nowadays malware only infects a few
level of knowledge and a larger amount of           hundred PCs before updating itself with a
time dedicated to each sample than                  new, undetectable variant to avoid detection
historically. Because of this situation antivirus   by regular antivirus signatures. The
engineers can no longer be employed “by the         underlying issue is how does an antivirus lab
numbers” to create hundreds of thousands of         become aware of such an infection if it is
signatures every few months.                        only affecting a handful of users?

2.2 Malware techniques                              2.2.2 Malware QA
and design                                          An older technique used incrementally by
                                                    malware today is basic QA testing. This is
The main differences between past viruses           done by testing each variant against the
and today’s malware is that the lifecycle           most common antivirus engines to make
has been significantly shortened and the            sure it goes undetected by the majority
objectives refined; to steal identities, use        of them.
computers as spam bots, steal online ban­
king credentials, credit card information,          This task is greatly simplified by online-
web logins, etc.                                    scanning services such as Jotti, VirusTotal,
                                                    the antivirus vendors’ online scanning
More importantly, today’s malware is                services 7 and online sandboxing services
designed to not raise any alarms. Unlike in         such as Cwsandbox, Norman and Anubis.
the past where viruses and worms were
designed to spread to as many computers             Malware creators also count on customized
as possible without user intervention,              tools to automate testing of new malware
generating a lot of noise and media                 against signatures, heuristics and even
awareness, today’s criminal malware wants           behavioral analysis technologies. With
to be as inconspicuous as possible. In order        these tools malware writers can test the
to achieve its objective, malware today uses        quality of their creations off-line, without
advanced techniques to evade detection              risking having the sample sent to the
and “fly low”.                                      antivirus laboratories via the above-
                                                    mentioned online scanning services.

2.2.1 Targeted Attacks: staying                     The objective of malware QA testing is not
below the radar                                     so much to avoid detection by all scanners
                                                    and all proactive techniques (generic signa­
One of the main strategies used by Targe­           tures, heuristics, behavior analysis, behavior
ted Attacks for staying below the radar is          blocking, etc.) but to avoid the majority of
to distribute few copies of many variants6.         them. Given its objective of staying below
FROM TRADITIONAL ANTIVIRUS TO COLLECTIVE INTELLIGENCE                                   Pag. 6

the radar it is not worth creating the most       has become a problem for antivirus
undetected malware if it is only going to li­     laboratories that approach malware reverse
ve for a few hours or days.                       engineering in a traditional manner and
                                                  need to analyze each sample one by one.

                                                  Not only the antivirus labs are having
2.2.3 Rootkits and sandbox                        problems with rootkits, but also companies
detection techniques                              are starting to experience the negative
                                                  effects of rootkits in business, especially
Another common detection evading techni­          when it is used for corporate espionage10.
que which is gaining momentum8 is the use of
rootkit techniques within Trojan and Spyware      In order to get a better idea of the problem
samples. When used by malware, rootkits           at real user’s machines we have gathered
create yet another barrier for being detected,    all known and unknown rootkit detections
especially as advanced rootkit detection tech­    by our free utility Panda Anti-Rootkit 1 1
nologies have not yet been deployed to all        between the months of December 2006
mass-production security solutions.               and June 2007 and mapped the
                                                  distribution of rootkits within malware in
It also means that the antivirus laboratories     the wild. The resulting “Top 5 rootkits in
need to spend more time analyzing kernel          the wild” are shown in figure 2 below,
mode drivers than user-mode samples. For          which shows a great increase in the use of
example LinkOptimizer, which has been             kernel-mode rootkits.
seen in-the-wild in recent months, is able
to determine if the machine it is about to
infect has security, debugging or system          2.2.4 Runtime-packers
monitoring tools installed. It also checks if
it is running in a Virtual Machine                Perhaps the most common technique to try
environment. If these checks are matched it       to evade detection by anti-malware pro­
silently exits and does nothing.                  ducts is the use of obscure runtime packers
                                                  with anti-debugging and anti-virtualization
Labs that depend on VM will have to go            techniques.
through great lengths to be able to install
certain LinkOptimizer samples in order to         These types of tools can modify and com­
analyze them in depth.                            press an executable file by encrypting and
                                                  changing its form from its original format.
At the time of writing few anti-malware           The final result is a modified executable
and security suites include some basic form       which, when executed, does exactly the sa­
of rootkit detection such as low-level            me thing as the original code, but from the
access cross-view against API-level calls, but    outside has a completely different form
most have not yet incorporated the more           and therefore evades signature-based de­
advanced rootkit detection and                    tection unless either the engine has the
deactivation techniques found in free,            specific unpacking algorithm or it is able to
stand-alone anti-rootkit utilities9.              unpack it generically.

Overall the use of rootkits by malware            Malware writers caught up to this
creators keeps steadily growing and this          approach and we are now even seeing
FROM TRADITIONAL ANTIVIRUS TO COLLECTIVE INTELLIGENCE                                                   Pag. 7

    Top Rootkit                         User-mode                Kernel-mode                 Prevalence

    1 Beagle.Fu                                                          x                      6.20%
    2 NaviPromo                                x                                                5.73%
    3 Rustock.A                                                          x                      1.20%
    4 Flush.K                                  x                                                1.01%
    5 Oddysee.B                                                          x                      0.20%
Figure 2: Top Rootkits in the Wild as detected by Panda Anti-Rootkit from December 2006 to June 2007

malware which use either modified                         2.2.5 Botnets
versions of known packers or even create
their own runtime packing routine                         According to some studies approximately
specifically for their malware samples 12 .               11% of computers worldwide are infected
                                                          by bots, which are responsible for sending
In order to address this problem, Panda’s                 up to 80% of all spam 14 . A large portion
engineers have created both generic packer                of money made by cyber criminals stems
detectors and generic unpacking                           from botnets.
algorithms which can detect unknown
packers and try to unpack them.                           The control of these large networks of
                                                          compromised machines is sold or rented to
However, a more effective solution will be                perform certain types of cyber-criminal
to at least flag the newly created runtime                activities, from sending spam runs,
packers as suspicious altogether. Some off-               distributed denial of service attacks, renting
the-shelf perimeter solutions already do                  of proxies, keylogging, pay per click installs,
this by default. Even some host-based                     adware installations, stored passwords,
security solutions are using this approach                man in the middle attacks, etc.
by flagging these types of samples as
malicious as is becoming obvious from the                 PandaLabs has witnessed on-line wars
different detection names used by the                     between different bot gangs to win over
different anti-malware engines13.                         hijacked PCs. Even though some evidence
                                                          suggests that there are many PCs
The impact of such an approach to                         belonging to Fortune 500 companies
proactive packer detection is not without                 which are controlled remotely by bot
cost. While speaking to other anti-malware                herders to send out spam 15 , the reality is
vendors during the 2007 International                     that virtually every corner of the Internet is
Antivirus Testing Workshop in Iceland it                  infested by bots.
became apparent that doing so in
corporate environments was a good                         Even though traditional botnets are
approach, but vendors with high install                   controlled via IRC, new P2P and HTTP
base on the consumer market could face                    based botnets which use stronger
such a high wave of false positives that the              communication encryption are becoming
solution could potentially be worse than                  popular among cyber-criminals in order to
the problem itself.                                       evade detection and shutdown.
FROM TRADITIONAL ANTIVIRUS TO COLLECTIVE INTELLIGENCE                                  Pag. 8

2.2.6 Staged infection vectors                    graphical tools emerge that simplify the
                                                  creation of new downloaders19, even with
It’s nothing new that most of today’s             custom packing techniques to evade
malware has a tendency of using a two-            detection.
staged attack as its main infection
technique, either by exploiting known or
zero-day vulnerabilities or by using small        2.2.7 “Malware 2.0”
downloaders which change very rapidly to
avoid detection.                                  A current trend in malware creation is that
                                                  the actual binary that infects the user’s PC
While in the past it would take malware           is “dumb” and the intelligence is “in-the-
authors weeks or even months to take              cloud”. The code that resides on the PC
advantage of a vulnerability as its main          has some simple functions that it passes on
infection vector, nowadays its normal to          to a remotely compromised server. The ser­
see exploits in the wild for vulnerabilities a    ver then returns instructions on what to do.
couple of days after it is known. Even            Borrowing the (perhaps overused) “2.0”
further, organizations that manage                term from current web trends, we will refer
darknets such as Team Cymru are seeing            to “Malware 2.0” as malware which sepa­
new zero-day exploits in the wild using           rates its intelligence from its code base.
stealthier techniques for days and weeks
before it is widely known and before they         PandaLabs has reported the “2.0”
are massively used by botnets.                    approach in banking targeted attack
                                                  Trojans in order to remotely monitor users’
Examples such as GDI, animated cursor and         browsing habits and, based on the online
VML vulnerabilities are being exploited by        banking landing page and authentication
automated infection frameworks such as            scheme, inject some type of HTML code or
Web-Attacker16 and MPack17, which make            other. Known banking Trojans such as
use of multiple vulnerabilities to exploit        Limbo/NetHell and Sinowal/Torpig use
unsuspecting and un-patched users in              these techniques quite extensively20
order to infect them with a Trojan.
                                                  Other “2.0” techniques recently used by
Downloaders have also become common               malware are “server-side-compilation”,
practice for two-staged infection                 where the webserver re-compiles a new
techniques. First a small file is executed        binary every few hours. Lastly, botnets are
either via a browser drive-by download or         using fast-flux DNS networks for improved
similar exploit. This file is coded with a        resistance against take-down efforts. These
single objective in mind; download a              last techniques are more visible in the
second file from a URL and execute it. This       recent Storm/Nuwar attacks.
second file in turn is the true Trojan which
ends up infecting the system.

These downloaders have become very
advanced. SecuriTeam recently ran a Code
Cruncher competition to create the
smallest downloader in the world18. More
recently we are seeing a myriad of
FROM TRADITIONAL ANTIVIRUS TO COLLECTIVE INTELLIGENCE                                            Pag. 9

3. Panda’s Technology Evolution

D                                                          3.1 First Generation:
        ealing with this malware situation
        using a traditional signature
        approach has not been valid for                    Antivirus
some years now. A complete Host Intrusion
Prevention System (HIPS) with advanced                     The first generation of antivirus products
heuristics, deep packet inspection firewall,               was purely based on signature detection.
behavior blocking, behavior analysis and                   This generation of technology occupied
system and application hardening are an                    most of the 1990’s and included polymor­
absolute must for any security solution,                   phic engines as well as basic rule-based
even though the sad reality is that about                  MS-DOS, Win32, Macro and, later on,
half the solutions on the market do not                    script heuristics. This period was also mar­
have these types of technologies yet. 2 1                  ked by the appearance of the first massive­
                                                           ly used win32 Trojans, such as NetBus and
At Panda we research and develop                           BackOrifice.
100% of our core anti-malware techno­
logies. This dedication to innovation has
allowed us to lead the way in proactive                    3.2 Second Generation:
technology deployment to the market.
                                                           Anti-malware
Following a defense-in-depth philosophy,
                                                           Starting in 2000 new types of malware star­
which could be summarized as integrating
                                                           ted to emerge, with file-less network worms
different protection technology layers at
                                                           and spyware taking the spotlight causing
different infrastructure layers, Panda
                                                           massive and highly visible epidemics.
Research, a team dedicated to developing
new security technologies, developed a
                                                           Basic antivirus engines evolved to integra­
new focus to security protection which is
                                                           te personal firewalls to be able to identify
based on the concept of Collective
                                                           and stop network worms based on packet
Intelligence.
                                                           signatures as well as system cleaners to
                                                           restore modified Operating System set­
The Collective Intelligence concept is
                                                           tings such as registry entries, HOST files,
d e s i g n e d t o c o m p l e m e n t P a n d a ’s
                                                           Browser Helper Objects, etc. It is within
integrated desktop, server and gateway
                                                           this second generation of technologies
protection to take the battle against today’s
                                                           that Panda Security integrated the Smart­
malware dynamic head on and provide the
                                                           Clean functionality into the anti-malware
f i n a l c o m p l e m e n t o f P a n d a ’s i d e a l
                                                           engine, designed to disinfect and restore
protection model.
                                                           the Operating System from a spyware or
                                                           Trojan backdoor infection.
Before we dive into explaining Collective
Intelligence, let’s do a walk-through of the
different technology generations on top of
which Collective Intelligence is built.
FROM TRADITIONAL ANTIVIRUS TO COLLECTIVE INTELLIGENCE                                   Pag. 10

3.3 Third Generation:                             havioral blocking, also known as system
                                                  and application hardening. Before going into
Proactive technologies                            each of these let’s take a look at the under­
                                                  lying uncloaking layer which makes malware
Panda released TruPrevent® behavioral tech­       visible to these behavioral technologies.
nologies in 2004 after more than three years
of intensive research and development.
                                                  3.3.1 Uncloaking techniques
Since then, TruPrevent® has evolved into a
set of behavioral technologies that are           As malware has evolved so have the
substantially more effective at blocking          techniques used to evade detection and
zero-day malware proactively without any          hide from prying eyes.
dependency on viral signatures than any
other previous effort in such direction.          To combat these hiding techniques there is
TruPrevent® is constantly adapted to new          an underlying layer of uncloaking
malware techniques and exploits.                  technologies common to all of Panda
                                                  products.
TruPrevent® was designed as an additional
protection layer to the anti-malware              The following techniques are able to
engine. Currently there are more than 5           inspect any item as deeply as necessary,
million computers running TruPrevent®. All        even if the item is making use of stealth
these computers also act as high-                 techniques to remain hidden in the system,
interaction honeypot nodes which report to        and pass on the results to the scanning and
PandaLabs any new malware sample that             monitoring technologies:
TruPrevent® flags as suspicious and which
is not detected by regular antivirus                    ·   Deep Code Inspection
signatures.                                             ·   Generic Unpacking
                                                        ·   Native File Access
TruPrevent's® approach consists of                      ·   Rootkit Heuristic
scanning each item or potential threat
using different techniques, carrying out in-
depth complementary inspections at the
different layers of the infrastructure. The       3.3.2 TruPrevent
approach to TruPrevent® implementations           Behavior Analysis
is modular and therefore can be applied
both to desktops and servers to become            Codenamed Proteus, it acts as a true last li­
full-blow integrated Host Intrusion               ne of defense against new malware execu­
Prevention Systems (HIPS).                        ting in the machine that manages to
                                                  bypass signatures, heuristics and behavior
As an approximate detail of its                   blocking. Proteus intercepts, during runti­
effectiveness, about two thirds of the new        me, the operations and API calls made by
malware samples received at PandaLabs             each program and correlates them before
from our users are now coming from                allowing the process to run completely. The
automated submissions from TruPrevent®.           real-time correlation results in processes
Technically TruPrevent® consists of 2 main        being allowed or denied execution based
technologies: behavioral analysis and be­         on their behavior alone.
FROM TRADITIONAL ANTIVIRUS TO COLLECTIVE INTELLIGENCE                                                   Pag. 11

As soon as a process is executed its              name. Several third-party tests have been
operations and API calls are monitored            performed on TruPrevent®.Performing
silently by Proteus, gathering information        tests for behavioral technologies such as
and intelligence about that process's             TruPrevent, using real-life malware
behavior. Proteus exhaustively analyses the       samples, is time-consuming and it requires
behavior and is designed to block the             a fair amount of expertise in the field. It is
malware as soon as it starts performing           without doubt much more challenging
malicious actions.                                than performing on-demand tests of
                                                  antivirus scanners against a collection of
If it is determined as suspicious, the process    viruses.
is blocked and killed before it can carry out
all of its actions and prevented from             The first test was commissioned by Panda
running again.                                    and it was performed by ICSALabs, a
                                                  Division of CyberTrust Corporation, in the
Unlike other behavioral technologies,             fall of 2004. ICSALabs tested the
Proteus is autonomous and does not                technologies against a set of approximately
present technical questions to the end user       100 real malware samples. This first test
("Do you want to allow process xyz to             was designed to verify that the
inject a thread into explorer.exe or memory       technologies worked against a variety of
address abc?"). If Proteus thinks that a          malware types, rather than to reach a
program is malicious it will block it without     conclusion about the overall effectiveness
requiring user intervention.                      of the technologies over time.

Most users cannot make informed                   At the same time, ICSALabs tested
decisions when it comes to security. Some         TruPrevent® against several sets of
behavioral products throw non-                    legitimate applications, from games to
deterministic opinions –or behavioral             Peer-to-peer packages, but was not able to
indecisions- whose effectiveness depends          produce any instance of false positives,
on the user clicking on the right choice. A       despite their efforts in this regard.
key functionality of any behavioral
technology must be making decisions               Another “early” review by PC Magazine
without user intervention. Anything less is       USA concluded that “TruPrevent blocked
a potential point of failure.                     two-thirds of a sample of recent worms,
                                                  v i r u s e s , a n d Tr o j a n s b a s e d s t r i c t l y o n
Our internal statistics show that this            behavior. Blocked no legitimate programs.
technology alone is capable of detecting          No noticeable impact on system
over 80 percent of the malware in the wild        performance.”
without signatures and without generating
false positives.

This technology does not require signature
updates, as it is based solely on the
behavior of applications. A bot would not
be a bot if it didn’t behave as such, but if it
does so it will be detected by this
technology, regardless of its shape or
FROM TRADITIONAL ANTIVIRUS TO COLLECTIVE INTELLIGENCE                                    Pag. 12

Figure 3: Panda’s integrated endpoint security

3.3.3 TruPrevent® Behavior                         wed and denied actions for a particular
                                                   application of group thereof. Rules can be
Blocking
                                                   set to control an application’s access to fi­
                                                   les, user accounts, registry, COM objects,
Codenamed KRE (Kernel Rules Engine), this
                                                   Windows services and network resources.
is TruPrevent’s second main component,
also known as Application Control &                Despite offering a high degree of
System Hardening or Resource Shielding.            granularity to administrators for creating
                                                   custom policies, the Application Control &
Hackers and malware abuse the privileges           System Hardening module (KRE) is shipped
of legitimate applications to attack systems       with a set of default configuration policies
by injecting code. To prevent these types of       which are managed and updated by
attacks generically it is very cost-effective to   PandaLabs.
use rule-based blocking technology which
can restrict the actions that authorized           The default policies provide protection
applications can perform in the system.            against attacks exploiting common
                                                   weaknesses found in out-of-the-box as well
KRE is composed of a set of policies which         as fully-patched installations of Windows
are defined by a set of rules describing allo­     operating systems.
FROM TRADITIONAL ANTIVIRUS TO COLLECTIVE INTELLIGENCE                                     Pag. 13

A recent example of the effectiveness that        algorithm. The genetic traits define the
proactive blocking provides is the never-         potential of the software to carry out
ending wave of Microsoft Office format            malicious or harmless actions when
vulnerabilities which are being exploited to      executed on a computer. GHE is capable of
hide malicious code24. These vulnerabilities      determining whether a file is innocuous,
have been used recently by targeted               worm, spyware, Trojan, virus, etc. by
attacks on certain companies. According to        correlating the different traits of each item
a study of known (patched) and zero-day           scanned.
(un-patched) Microsoft Office vulnerability
exploits, an average antivirus signature          GHE can be set to low, medium or high
detection rate of 50% was achieve by all          sensitivity with the obvious combination
tested antivirus engines. That’s a one-in-        trade-off between detection rates and false
two chance of being infected by simply            positives. The different sensitivity levels are
opening an exploited Microsoft Word,              designed to be applied to different
PowerPoint or Excel document.                     environments depending on the probability
                                                  of malware prevalence on each.
On the contrary, behavioral blocking
technologies such as TruPrevent,                  For example at network SMTP gateways we
proactively prevents Microsoft Word,              have found that the likelihood of an
PowerPoint, Excel, Access, Acrobat Reader,        executable files being malware is very high.
Windows Media Player and other                    Therefore the implementation we have
                                                  done in our commercial products is of high
applications from dropping and running
                                                  sensitivity for network layer e-mail scanning
any type of executable code on the system.
                                                  products. However for storage (or
Unlike any antivirus signatures tested,
                                                  application) layers where the vast majority
TruPrevent® provides real zero-day
                                                  of executable code is from legitimate
protection against any Microsoft Office
                                                  applications, we have implemented GHE
exploit, known or unknown.
                                                  with medium sensitivity. With this setting
                                                  we've been able to maximize detection
                                                  rates for unknown malware while having a
3.3.4 Genetic Heuristics                          negligible false positive rate.

“Genetic” technologies are inspired by the        The results of the GHE so far are excellent.
field of genetics in biology and its useful­      Since its release, roughly one third
ness to understand how organisms are in­          (cumulative) of the new variants received at
dividually identified and associated to other     PandaLabs from real users' machines have
organisms. These technologies are based           been submitted automatically by the GHE.
on the processing and interpretation of
"digital genes", which are represented in
our case by quite a few hundred characte­
ristics of each file that is scanned.

Codenamed Nereus, the Genetic Heuristic
Engine was initially released in 2005. The
objective of GHE is to correlate the genetic
traits of files by using a proprietary
FROM TRADITIONAL ANTIVIRUS TO COLLECTIVE INTELLIGENCE                                   Pag. 14

3.4 Collective Intelligence                       3.4.1 Benefiting from
                                                  Community Knowledge
Today there is over 10 times more malware
being distributed than two years ago. The         Traditional security solutions are architected
obvious conclusion is that a security             with a PC-centric philosophy. This means
solution must detect 10 times more                that a PC is treated as a single unit in time
malware to provide adequate protection to         and any malware detected within that PC is
users. While a full-fledged HIPS solution         considered separately from the rest of the
raises the bar substantially by detecting and     malware samples detected in millions of
blocking most of these with proactive             other PCs.
technologies, it is still possible for unknown
malware to slip through its defenses. We          Traditional security companies do not have
need to consider the fact that, while 80%         visibility into what PC a particular piece of
or 90% of proactive effectiveness is              malware was first seen on. Neither is there
relatively speaking an excellent score, in        visibility of the continuity of that malware’s
absolute terms it may lead to hundreds or         evolution over time in different PCs.
thousands of malware samples being
missed over time, since even a small              Most importantly, other PCs do not
fraction of a large enough number will still      automatically benefit of proactive malware
be a “big” number.                                detections on different PCs. They have to
                                                  wait for the antivirus lab to receive that
The Collective Intelligence approach is           specific sample, wait for a signature to be
initially released at the end of 2006 in          created, QA’ed, deployed and protect other
limited pilots with the objective of being        users.
able to reliably detect “10 times more than
we are currently detecting with 10 times          Ultimately this results in traditional
less effort”. Collective Intelligence             approaches being too slow to combat
functions as an online and real-time              today’s rapidly moving malware.
Security-as-a-Service (SaaS) platform. With
over two years of research and                    One of the main benefits of the Collective
development behind it and millions of             Intelligence approach, in addition the
dollars in investment efforts, it is already      effectiveness provided by the automation
paying off by:                                    of the malware remediation life-cycle, is the
                                                  automatic and real-time benefit it provides
1. Benefiting from “community”                    to the users of the Collective Intelligence
knowledge to proactively protect                  Community.
others.
2. Automating and enhancing malware               As soon as a malicious process is detected
collection, classification and                    in a users’ PC by the Collective Intelligence
remediation.                                      servers (whether by system heuristics,
                                                  emulation, sandboxing or behavioral
3. Gaining knowledge on techniques to
                                                  analysis, etc.) the rest of the users
improve existing technologies.
                                                  worldwide will automatically benefit in real-
4. Deploying new generation of                    time from that specific detection. This
security services from the cloud.                 results in a close to real-time detection not
FROM TRADITIONAL ANTIVIRUS TO COLLECTIVE INTELLIGENCE                                   Pag. 15

only of initial malware outbreaks but also        the vast majority of samples. Let’s walk
of targeted attacks whose objective is            through the process from the point of view
infecting a small number of users to stay         of a computer who has just been exploited
below the radar.                                  and infected by a malicious code.

3.4.2 Automated Malware                           3.4.2.1 Automated malware
Protection Process                                collection

One of the biggest barriers to raising the        The Collective Intelligence (CI) agent
bar of reliable malware detection ratios is       gathers information of processes and
the fact that the process of creating a           memory objects and performs queries
signature against a single sample takes too       against the CI central servers which
long in the industry. Each malware sample         perform a variety of checks against those.
needs to be sent to the lab by an affected
user or fellow researcher, reversed               If certain conditions are met, the suspicious
engineered by a lab technician which in           file or parts thereof is automatically
turn needs to create a detection signature        uploaded, with the users consent, to the CI
and disinfection routine for it. These in turn    servers where it is further processed.
need to be quality-assured, uploaded to
production servers, replicated worldwide          Since processes loaded in memory are not
and finally downloaded and applied by             subject to many of the cloaking techniques
customers.                                        and “reveal themselves”, the agent
                                                  component does not need to contain a
This entire process is, in most cases, mostly     large amount of intelligence and
manual and can take up anywhere from              uncloaking routines and can therefore be
minutes, to hours or days or even weeks,          very light.
depending on the workload of the lab
engineers and other factors such as sample        Panda has built a vast database of malware
priority, prevalence, damage potential,           samples, which are automatically collected,
media coverage, etc.                              which in turn provides the CI web-service
                                                  with a real-time feed of new malware
The process can be even delayed much              classification entries.
longer when “intelligence” or functionality
upgrades to the anti-malware or behavioral
engines are involved. It is typical of an anti-
malware vendor to upgrade its solutions
                                                  3.4.2.2 Automated malware
once or twice a year, as each upgrade has         classification
a costly testing and deployment process for
corporate customers.                              Server-based processing is not limited by
                                                  the CPU and memory constraints of
Thanks to the Collective Intelligence             personal computers. Therefore scanning
infrastructure this entire process of malware     routines at the CI servers undergo much
collection, classification and remediation        more in-depth processing by more sensitive
can be automated and performed online for         technologies (signature and sensitive
FROM TRADITIONAL ANTIVIRUS TO COLLECTIVE INTELLIGENCE                                    Pag. 16

heuristics scanning, emulation, sandboxing,       which have been gradually deployed to our
virtualization, white-listing, etc.) to reach a   existing products.
final classification.
                                                  One of the main benefits of the Collective
It is important to note that the scanning         Intelligence approach is that these
power used at the CI servers is only limited      signatures do not need to be downloaded
by hardware and bandwidth scaling, unlike         to each client as they operate from the
a typical scenario at a PC, desktop or server     cloud. This however does not mean that
machine. Therefore many of the more               the client machine will not need to
resource-intensive proactive techniques           maintain updated signatures.
which PandaLabs is using, and which
provide much higher detection rates (at an        A potential threat to such an approach is
also higher computational costs) can now          the availability of the Collective Intelligence
be used massively for the benefit of the          servers. However our approach for
users without even touching valuable              integration of the Collective Intelligence
customer’s CPU and memory resources.              technology on current solutions is designed
                                                  as an additional layer of protection.
With this approach the majority of new            Therefore under non-availability of the
malware samples can be analyzed and               platform for whatever reason, security
classified automatically in a matter of           protection would fall back to the regular
minutes. The CI servers are managed by            HIPS solution which provides well above
PandaLabs and therefore samples that              average protection.
cannot be classified automatically are
ultimately looked at by an analyst at the lab.
                                                  3.4.3 Gaining Knowledge on
                                                  Malware Techniques
3.4.2.3 Automated malware
remediation                                       Other one of the main benefits provided by
                                                  the community feature of Collective Intelli­
The remediation module of the CI is in            gence is that of giving insight to our engi­
charge of automatically creating detection        neers of new malware techniques and dis­
and disinfection signatures for the samples       tribution points. Questions such as where
previously analyzed by the processing and         was a specific piece of malware first found
classification module. These signatures are       and how did it spread allow us to model
in turn used by the community of CI users         additional intelligence into specific malware
to proactively detect and disinfect new or        families and even creators of specific
even targeted attacks with very low num­          malware variants.
bers of infected hosts.
                                                  This approach of applying data
The traditional anti-malware and HIPS             warehousing and data mining techniques
solutions have also started to benefit from       to malware detections by the community
the CI approach. During the initial 3             provides significant knowledge on how
months of operation the remediation               malware and targeted attacks are carried
module has created protection for a few           out. The type of knowledge that can be
hundreds of thousands of malware samples          gathered using this approach becomes
FROM TRADITIONAL ANTIVIRUS TO COLLECTIVE INTELLIGENCE                                                 Pag. 17

especially useful if it can be applied for        Initially this approach might have seemed
tracking infection origins, which in turn         as an interesting idea back then. However
might have some interesting applications          the challenges presented by a white-listing
and benefits for law enforcement efforts.         solution to completely prevent malware are
                                                  varied. Some of the main shortcomings of
                                                  a “white-listing only” approach are:
3.4.4 Deploying Security Servi­
                                                  1. There are billions of goodware files vs.
ces “from-the-cloud”                                 the few millions of malware files in
                                                     existence today. For white-listing to be
We have developed and deployed a few
                                                     effective you would have to analyze
services already that function purely based
                                                     many more files than malware.
on the Collective Intelligence platform. The­
se online services are designed to perform
                                                  2. Every time a new file has to be added to
in-depth audits of machines and detect
                                                     the white-list, it needs to be analyzed to
malware not detected by the installed secu­
                                                     make sure it is not malicious. Simply
rity solution.
                                                     adding files to the white-list without
                                                     analyzing them completely defeats the
For consumers and stand-alone PCs we
                                                     purpose of a white-list. Otherwise how
h a v e d e p l o y e d N a n o S c a n              do you prevent malware from being
(www.nanoscan.com) which scans a PC for              included on the white-list?
malware actively running and TotalScan
(www.pandasecurity.com/totalscan) which           3. Every time a new update or upgrade is
performs a full system scan of the entire            made available as a Service Pack or
PC, including hard drive, memory, email              Hotfix for Microsoft Windows, Office,
databases, etc.                                      QuickTime, Adobe, Java, etc. the white-
                                                     listed files need to be re-analyzed and
On the corporate front the requirements for          re-created.
performing and in-depth malware audit are
more demanding. Therefore we have                 4. Managing these white-lists on each
created a specific managed service called            computer on a network is a manual and
Malware Radar (www.malwareradar.com).                tedious job which needs to be done and
Thanks to this service companies can quickly         which network administrators need to
perform complete audits of their entire              find time to perform.
network endpoints to verify their level of
security, pinpoint non-detected infection         5. I f a n t i v i r u s l a b o r a t o r i e s w h o h a v e
sources or to unveil machines which have             hundreds of engineering resources
been subject to targeted attacks.                    cannot keep up with the pace of
                                                     analyzing all the malware, how much
                                                     investment in capital and resources
3.4.5 A note on white-listing                        should a white-listing company require
                                                     in order to keep up with the pace of
Since 2004 there have been some new                  analyzing 100 times more goodware?
companies spawn from under the rocks
promising to “get rid of the virus problem        6. Anti-malware updates are delivered to
forever” with a white-listing approach.              customers via signature databases which
FROM TRADITIONAL ANTIVIRUS TO COLLECTIVE INTELLIGENCE                                  Pag. 18

   are already big in size. However white-        white-listing component is an important
   listing updates will be much bigger in si­     aspect for complementing and improving
   ze. How shall those be delivered to the        black-list detection and, specially, reducing
   desktop and companies?                         false positives and processing times.

7. What happens when there’s new or
   updated applications that a user or
   company needs to run which are not
   included in the white-list? Who will be
   doing the reverse engineering and
   analysis of the supposedly benign
   program and associated files to
   determine that they are truly non-
   malicious?

8. What happens when a virus or worm
   manages to infect files of a white-listed
   reputable software company’s installer
   package? It has happened in the past a
   few times already.

Relying exclusively on white-listing
technologies might make sense in certain
locked down environments such as call
centers, ATM machines and the like. But in
the vast majority of corporate environments
this is not the case.

There have been very active and lively
discussions25 26 lately on the pros and cons
of white-listing, specially promoted by the
white-listing companies themselves that
feed on the “Antivirus is Dead, White-
Listing is the solution” rumor.

However the white-listing approach should
not be dismissed altogether. It does bring
up many interesting opportunities in the
fight against malware, but we believe the
benefits are much more effective when
combined with black-listing and other
proactive approaches.

As we have seen during the explanation of
the Collective Intelligence platform, a
FROM TRADITIONAL ANTIVIRUS TO COLLECTIVE INTELLIGENCE                                   Pag. 19

4. Conclusion

T
        he latest advances by the black hat and cybercrime communities are taking
        advantage of the inherent weaknesses in the security industry: (a) the labs are being
        swamped by more malware which is being created every day, (b) by remaining
invisible users do not perceive the need for additional protection, (c) targeted attacks that
only infect very few users are more effective than epidemic attacks that infect millions of
users and (d) users tend to trust a single solution or single layer of protection as their main
line of defense against malware.

As malware techniques advance in this cat-and-mouse game, security vendors need to add
more layers of protection to keep customers safe. The need for additional protection is
revealed by the fact that a large portion of users with current and updated security
solutions is in fact infected.

To tackle today’s problem we need new layers of protection that take advantage of
automating the entire malware protection cycle, from sample collection, analysis,
classification to remediation. But automation by itself is not enough. We also need visibility
into what’s happening on all PCs in order to detect targeted attacks more efficiently and
gain a competitive edge on malware creators.

The approach developed by Panda Security, called Collective Intelligence, provides all the
benefits of an added layer of defense that provides effective response and protection to
the current malware threats, is able to detect targeted attacks and gains intelligence
thanks to the correlation of all the detections by the community of users.
FROM TRADITIONAL ANTIVIRUS TO COLLECTIVE INTELLIGENCE                                                                Pag. 20

5. References

1. Research Study: Active Infections in Systems Protected by       17. MPack Uncovered, May 2007.
Updated AntiMalware Solutions. Panda Research. August              http://blogs.pandasoftware.com/blogs/images/PandaLabs/2
2007.                                                              007/05/11/MPack.pdf
http://research.pandasecurity.com
                                                                   18. The World’s Smallest Downloader. Symantec. December
2. Gartner's 10 Key Predictions for 2007. Gartner.                 2006.
December 2006.                                                     http://www.symantec.com/enterprise/security_response/we
http://www.eweek.com/article2/0,1895,2072416,00.asp                blog/2006/12/worlds_smallest_downloader.html

3. The Zero-Day Dilemma. Security IT Hub. January 2007.            19. Packing a punch (II). Panda Research. March 2007.
http://www.security.ithub.com/article/The+ZeroDay+Dilem            http://research.pandasoftware.com/blogs/research/archive/2
ma/199418_1.aspx                                                   007/03/20/Packing-a-Punch-_2800_III_2900_.aspx

4. Welcome to 2007: the year of professional organized             20. Banking Targeted Attack Techniques. Panda Research.
malware development. F-Prot’s Michael St. Neitzel at Hispasec.     March 2007.
February 2007.                                                     http://research.pandasoftware.com/blogs/images/Panda-
http://blog.hispasec.com/virustotal/16                             eCrime2007.pdf

5. Call the cops: We're not winning against cybercriminals.        21. Host-Based Intrusion Prevention Systems (HIPS) Update:
ComputerWorld. February 2007.                                      Why Antivirus and Personal Firewall Technologies Aren't
http://www.computerworld.com/action/article.do?comman              Enough. Gartner. January 2007.
d=viewArticleBasic&articleId=9010041                               http://www.gartner.com/teleconferences/attributes/attr_165
                                                                   281_115.pdf
6. The Long Tail: malware’s business model. Panda Research.
January 2007.                                                      22. A Very Large Honeynet. Panda Research. December 2006.
http://research.pandasoftware.com/blogs/research/archive/2         http://research.pandasoftware.com/blogs/research/archive/2
007/01/08/The-Long-Tail_3A00_-malware_2700_s-business-             006/12/19/A-very-large-malware-honeynet.aspx
model.aspx
                                                                   23. Panda TruPrevent Personal 2005. PC Magazine USA.
7. List of online scanners. CastleCops Wiki.                       November 2004.
http://wiki.castlecops.com/Online_antivirus_scans                  http://www.pcmag.com/article2/0,1759,1727653,00.asp

8. Kernel Malware. F-Secure. February 2007.                        24. The Last Great Security Crisis. eWeek February 2007.
http://www.f-secure.com/weblog/archives/archive-                   http://www.eweek.com/article2/0,1895,2095118,00.asp
022007.html#00001118

9. Antirootkit.com List of Rootkit Detection & Removal Software.   25. Comments on “The Decline of Antivirus and the Rise of
http://www.antirootkit.com/software/index.htm                      White-Listing”. The Register. June 2007.
                                                                   http://www.theregister.co.uk/2007/06/27/whitelisting_v_an
10. Rootkit used in Vodafone Phone Tapping Affair. July 2007.      tivirus/comments/
http://www.antirootkit.com/blog/2007/07/12/rootkit-used-
in-vodafone-phone-tapping-affair/                                  26. “More on White-listing”. Kurt Wismer. June 2007.
                                                                   http://anti-virus rants.blogspot.com/2007/06/more-on-
11. Panda Anti-Rootkit. April 2007.                                whitelisting.html
http://research.pandasoftware.com/blogs/research/archive/2
007/04/27/New-Panda-Anti_2D00_Rootkit-_2D00_-Version-
1.07.aspx

12. Packing a punch. Panda Research. February 2007.
http://research.pandasoftware.com/blogs/research/archive/2
007/02/12/Packing-a-punch.aspx

13. AV performance statistics. OITC & MIRT. Real-time feed
of antivirus zero-day detection.
http://winnow.oitc.com/avcentral.html

14. Attack of the Zombie Computers Is Growing Threat. The
New York Times. January 2007.
http://www.nytimes.com/2007/01/07/technology/07net.ht
ml?ex=1325826000&en=cd1e2d4c0cd20448&ei=5090

15. 30 Days of Bots Inside the Perimeter. Support Intelligence.
March-April 2007.
http://blog.support-intelligence.com

16. Web-Attacker Exposed. Websense. November 2006.
http://www.websense.com/securitylabs/blog/blog.php?BlogID=94
PANDA SECURITY
Panda SPAIN                         Panda USA
Ronda de Poniente, 17               230 N. Maryland, Suite 303
28760. Tres Cantos. Madrid. SPAIN   P.O. Box10578. Glendale, CA 91209 - USA
Phone: +34 91 806 37 00             Phone: +1 (818) 5436 901

www.pandasecurity.com

                                                          © Panda 2007. All rights reserved. 0907-WP-PSD-I-01

                                                                          www.pandasecurity.com
You can also read