Partner Perspectives // bitdefender
1/17/2017
12:00 PM
Liviu Arsene
Liviu Arsene
Partner Perspectives
Connect Directly
Twitter
Google+
LinkedIn
RSS
100%
0%

Machine Learning For Cybersecurity Not Cybercrime

Cybercriminals have yet to adopt machine learning for offensive attack strategies - and they probably won't for a long time.

The cybersecurity industry has always been under constant strain from cybercriminals and malware. With increasing integration of hardware, software and services being built into every aspect of our lives, the task of keeping data secure has become even more difficult.

The arsenal of tools that cybercriminals now have at their disposal has raised concerns for security companies, and turned the criminals into threat actors who can create, disseminate and penetrate a target’s defenses using custom-built and never-before-seen malware. The security industry has had to adopt a new way of dealing with the unknown by leveraging the powerful capabilities of machine learning algorithms.

Cybersecurity & Machine Learning
Because targeted and advanced threats that seek to prey on organizations and businesses often evade traditional security mechanisms, machine learning algorithms have stepped in to fill in the gap between proactivity and detection. While humans are great at in-depth analysis and pinpointing code subtleties in malicious samples, machine learning is better at applying models on large data without tiring or complaining of repetitive tasks.

In the context of big data – where everything connected to the Internet from IoT devices to physical and virtual endpoints is a potential source of information or point of attack - machine learning can be trained to parse, analyze and interpret that data with little no effort.

The human component, however, is responsible for the accuracy of the machine learning model and for supplying its “wits.” Cybersecurity specialists with years of experience in reverse engineering malware samples and analyzing attack techniques are the ones who usually transfer their experience to machine learning algorithms, training the algorithms for behavior analytics and anomaly detection. While machine learning algorithms range from neural networks to genetic algorithms, their ultimate goal is to adapt to variations of a baseline behavior.

Do Cybercriminals Use Machine Learning?
No, they don’t! That’s because they already have a wide range of tools and mechanisms that have automated not only malware development but also ensured that each new malware sample is unique.

Obfuscation and polymorphism are just two examples cybercriminals use to create and deliver ransomware samples to both average users and organizations. They are so effective that ransomware is estimated to have inflicted at least $1 billion in financial losses in 2016 alone.

Encryption is another powerful tool consistently leveraged by cybercriminals to mask data exfiltration and even extort victims. The whole point of the cybercrime industry is to constantly create new packing mechanisms for malware samples, and not necessarily come up with innovative attack techniques or behavior. This doesn’t require machine learning; it involves constant algorithm tweaking or the development of obfuscation functions or encryption algorithms.

Is Machine Learning Offensive or Defensive?
When applied in the “cyber” context, current machine learning capabilities are mostly defensive. Machine learning helps the security industry tackle more than 500 million malware samples. Cybercriminals have yet to adopt machine learning and they probably won’t for a long time.

While there have been examples of machine learning algorithms being pitted against each other; one looking for software vulnerabilities and the other trying to patch them – these exercises were for demonstration only.

Of course, machine learning can be considered to have offensive capabilities when applied in the gaming industry, as it can be trained to take out virtual foes with the same accuracy as their human counterparts. However, they’re yet to be used for cybercriminal activities. 

Liviu Arsene is a senior e-threat analyst for Bitdefender, with a strong background in security and technology. Reporting on global trends and developments in computer security, he writes about malware outbreaks and security incidents while coordinating with technical and ... View Full Bio
Comment  | 
Print  | 
More Insights
Comments
Newest First  |  Oldest First  |  Threaded View
Moises Danziger
50%
50%
Moises Danziger,
User Rank: Apprentice
3/24/2017 | 5:27:15 PM
I wouldn't be so sure
During my researchs in the application of Machine Learning (ML) on security I've been surprising with some proposals showing ways to apply ML to cybercrime. For example, for malware obfuscation, to improve the C&C channel from botnets, to add intelligence for worms...etc. These are simple examples. After that, I've been studying the impact of ML when in bad hands. I can tell you hackers will use ML to improve their attack tools soon (although they may already be applying). Our challenge is discovering how it could be dangerous for current security tools.
Register for Dark Reading Newsletters
Dark Reading Live EVENTS
INsecurity - For the Defenders of Enterprise Security
A Dark Reading Conference
While red team conferences focus primarily on new vulnerabilities and security researchers, INsecurity puts security execution, protection, and operations center stage. The primary speakers will be CISOs and leaders in security defense; the blue team will be the focus.
White Papers
Video
Cartoon Contest
Write a Caption, Win a Starbucks Card! Click Here
Latest Comment: "Jamie, the darn Unicorn is back."
Current Issue
Security Vulnerabilities: The Next Wave
Just when you thought it was safe, researchers have unveiled a new round of IT security flaws. Is your enterprise ready?
Flash Poll
Slideshows
Twitter Feed
Dark Reading - Bug Report
Bug Report
Enterprise Vulnerabilities
From DHS/US-CERT's National Vulnerability Database
CVE-2017-0290
Published: 2017-05-09
NScript in mpengine in Microsoft Malware Protection Engine with Engine Version before 1.1.13704.0, as used in Windows Defender and other products, allows remote attackers to execute arbitrary code or cause a denial of service (type confusion and application crash) via crafted JavaScript code within ...

CVE-2016-10369
Published: 2017-05-08
unixsocket.c in lxterminal through 0.3.0 insecurely uses /tmp for a socket file, allowing a local user to cause a denial of service (preventing terminal launch), or possibly have other impact (bypassing terminal access control).

CVE-2016-8202
Published: 2017-05-08
A privilege escalation vulnerability in Brocade Fibre Channel SAN products running Brocade Fabric OS (FOS) releases earlier than v7.4.1d and v8.0.1b could allow an authenticated attacker to elevate the privileges of user accounts accessing the system via command line interface. With affected version...

CVE-2016-8209
Published: 2017-05-08
Improper checks for unusual or exceptional conditions in Brocade NetIron 05.8.00 and later releases up to and including 06.1.00, when the Management Module is continuously scanned on port 22, may allow attackers to cause a denial of service (crash and reload) of the management module.

CVE-2017-0890
Published: 2017-05-08
Nextcloud Server before 11.0.3 is vulnerable to an inadequate escaping leading to a XSS vulnerability in the search module. To be exploitable a user has to write or paste malicious content into the search dialogue.