Retrospective / Proactive Test 2014
Heuristic and behavioural protection against new/unknown malicious software
|Test Period||March 2014|
|Number of Testcases||1200|
|Online with cloud connectivity|
|False Alarm Test included|
Many new malware samples appear every day, which is why it is important that antivirus products not only provide new updates, as frequently and as quickly as possible, but also that they are able to detect such threats in advance with generic/heuristic techniques; failing that, with behavioural protection measures. Even if nowadays most antivirus products provide daily, hourly or cloud updates, without proactive methods there is always a time-frame where the user is not reliably protected. The aim of this test is to evaluate the proactive detection and protection rates in this time-frame (without cloud). The data shows how good the proactive heuristic/generic detection and behavioural protection capabilities of the scanners were in detecting new threats used in this test. The design and scope of the test mean that only the heuristic/generic detection capability and behavioural protection capabilities were tested (offline). Additional protection technologies (which are dependent on cloud-connectivity) and infection vectors are considered by AV-Comparatives in e.g. Whole-Product Dynamic (“Real-World”) Protection Tests and other tests, but are outside the scope of the Retrospective/Proactive Tests.
This test report is the second part of the March 2014 test. The report is delivered several months later due to the large amount of work required, deeper analysis, preparation and dynamic execution of the retrospective test-set. This type of test is performed only once a year and includes a behavioural protection element, where any malware samples are executed, and the results observed. Although it is a lot of work, we usually receive good feedback from various vendors, as this type of test allows them to find bugs and areas for improvement in the behavioural routines (as this test evaluates specifically the proactive heuristic and behavioural protection components).
The products used the same updates and signatures that they had on the 7th March 2014. This test shows the proactive protection capabilities that the products had at that time. We used 1,200 new, unique and very prevalent malware samples that appeared for the first time shortly after the freezing date. The size of the test-set has also been reduced to a smaller set containing only one unique sample per variant, in order to enable vendors to peer-review our results in a timely manner. The following products were tested:
What about the cloud? Even several weeks later, a number of the malware samples used were still not detected by some cloud-dependent products, even when their cloud-based features were available. Consequently, we consider it a marketing excuse if retrospective tests – which test the proactive protection against new malware – are criticized for not being allowed to use cloud resources. This is especially true considering that in many corporate environments the cloud connection is disabled by the company policy, and the detection of new malware coming into the company often has to be provided (or is supposed to be provided) by other product features. Cloud features are very (economically) convenient for security software vendors and allow the collection and processing of large amounts of metadata. However, in most cases (not all) they still rely on blacklisting known malware, i.e. if a file is completely new/unknown, the cloud will usually not be able to determine if it is good or malicious.
We included in the retrospective test-set only new malware that was very prevalent in-the-field shortly after the freezing date. Samples which were not detected by the heuristic/generic detection capabilities of the products were then executed in order to see if behaviour-blocking features would stop them. In several cases, we observed that behaviour blockers only warned about some dropped malware components or system changes, without protecting against all the malicious actions performed by the malware; such cases were not counted as a block. As behaviour blockers only come into play after the malware is executed, a certain risk of being compromised remains (even when the security product claims to have blocked/removed the threat). Therefore, it is preferable that malware be detected before it is executed, by e.g. the on-access scanner using heuristics. This is why behaviour blockers should be considered a complement to the other features of a security product (multi-layer protection), and not a replacement.
The awards are given by the testers after consulting a number of statistical methods, including hierarchical clustering. We based our decisions on the following scheme:
|None - Few FPs|
|Very many FPs|
|Crazy many FPs|
To know how these antivirus products perform with updated signatures and cloud connection against prevalent malware files, please have a look at our File Detection Tests of March and September. To find out about real-life online protection rates provided by the various products, please have a look at our ongoing Whole-Product Dynamic “Real-World” Protection tests. Readers should look at the results and decide on the best product for them based on their individual needs. For example, laptop users who are worried about infection from e.g. infected flash drives whilst offline should pay particular attention to this Proactive test.
False Positive (False Alarm) Test Result
To better evaluate the proactive detection capabilities, the false-alarm rate has to be taken into account too. A false alarm (or false positive [FP]) occurs when an antivirus product flags an innocent file as infected. False alarms can sometimes cause as much trouble as real infections.
The false-alarm test results were already included in the March test report. For details, please read the report, available at http://www.av-comparatives.org/wp-content/uploads/2014/04/avc_fps_201403_en.pdf
|1.||ESET||1||very few FPs|
|3.||Bitdefender, BullGuard, Emsisoft, Lavasoft, Tencent||3|
|6.||AVG, Kaspersky Lab||9|
A small behavioural false-alarm test using 200 most downloaded/common software packages released in February did not bring up any additional false alarms.
The results show the proactive (generic/heuristic/behavioural) protection capabilities of the various products against new malware. The percentages are rounded to the nearest whole number.
Below you can see the proactive protection results over our set of new and prevalent malware files/families appeared in-the-field (1,200 malware samples):
|Blocked||User dependent||Compromised||Proactive / Protection Rate||False Alarms||Cluster|
 User-dependent cases were given a half credit. Example: if a program blocks 80% of malware by itself, plus another 20% user-dependent, we give it 90% altogether, i.e. 80% + (20% x 0.5).
Award levels reached in this Heuristic / Behavioural Test
The following awards are for the results reached in the proactive/behavioural test, considering not only the protection rates against new malware, but also the false alarm rates:
This test is an optional part of our public main test-series, that is to say, manufacturers can decide at the beginning of the year whether they want their respective products to be included in the test. The test is currently done as part of the public main-test series only if a minimum number of vendors choose to participate in it.
Microsoft security products are not included in the awards page, as their out-of-box protection is (optionally) included in the operating system and is currently considered out-of-competition.
Readers may be interested to see a summary and commentary of our test methodology which was published by PC Mag two years ago: http://securitywatch.pcmag.com/security-software/315053-can-your-antivirus-handle-a-zero-day-malware-attack
Copyright and Disclaimer
This publication is Copyright © 2014 by AV-Comparatives ®. Any use of the results, etc. in whole or in part, is ONLY permitted after the explicit written agreement of the management board of AV-Comparatives prior to any publication. AV-Comparatives and its testers cannot be held liable for any damage or loss, which might occur as result of, or in connection with, the use of the information provided in this paper. We take every possible care to ensure the correctness of the basic data, but a liability for the correctness of the test results cannot be taken by any representative of AV-Comparatives. We do not give any guarantee of the correctness, completeness, or suitability for a specific purpose of any of the information/content provided at any given time. No one else involved in creating, producing or delivering test results shall be liable for any indirect, special or consequential damage, or loss of profits, arising out of, or related to, the use or inability to use, the services provided by the website, test documents or any related data.
For more information about AV-Comparatives and the testing methodologies, please visit our website.