Enhanced Real-World Test 2020 – Enterprise
Advanced Threat Protection – Targeted Attacks, Exploits and Fileless Threats
|Test Period||September - November 2020|
|Number of Testcases||15|
|Online with cloud connectivity|
|False Alarm Test included|
“Advanced persistent threat” is a term commonly used to describe a targeted cyber-attack that employs a complex set of methods and techniques to penetrate information system(s). Different aims of such attacks could be stealing / substituting / damaging confidential information, or establishing sabotage capabilities, the last of which could lead to financial and reputational damage of the targeted organisations. Such attacks are very purposeful, and usually involve highly specialized tools. The tools employed include heavily obfuscated malicious code, the malicious use of benign system tools, and non-file-based malicious code.
In our Advanced Threat Protection Test (Enhanced Real-World Test), we use hacking and penetration techniques that allow attackers to access internal computer systems. These attacks can be broken down into Lockheed Martin’s Cybersecurity Kill Chain, and seven distinct phases – each with unique IOCs (Indicators of Compromise) for the victims. All our tests use a subset of the TTP (Tactics, Techniques, Procedures) listed in the MITRE ATT&CK framework. A false alarm test is also included in the report.
To represent the targeted system, we use fully patched 64-bit Windows 10 systems, each with a different AV product installed. In the enterprise test, the target user has a standard user account. In the consumer test, an admin account is targeted. For this reason and others (e.g. possibly different settings), the results of the Consumer Test should not be compared with those of the Enterprise Test.
Once the payload is executed by the victim, a Command and Control Channel (C2) to the attacker’s system is opened. For this to happen, a listener has to be running on the attacker’s side. For example, this could be a Metasploit Listener on a Kali Linux system. Using the C2 channel, the attacker has full access to the compromised system. The functionality and stability of this established access is verified in each test-case.
The test consists of 15 different attacks. It currently focuses on protection, not on detection, and is carried out completely manually. Whilst the testing procedure is necessarily complex, we have used a fairly simple description of it in this report. This is in accordance with reader feedback, and we hope that it will make it comprehensible to a wider audience.
We congratulate all those vendors who took part in the test, even those whose products did not get the best scores, as they are striving to make their software better.
Scope of the test
The Advanced Threat Protection (ATP) Test looks at how well the tested products protect against very specific targeted attack methods. It does not consider the overall security provided by each program, or how well it protects the system against malware downloaded from the Internet or introduced via USB devices.
It should be considered as a complement to the Real-World Protection Test and Malware Protection Test, not a replacement for either of these. Consequently, readers should also consider the results of other tests in our Main-Test Series when evaluating the overall protection provided by any individual product. This test focuses on whether the security products protect against specific attack/exploitation techniques used in advanced persistent threats. Readers who are concerned about such attacks should consider the products participating in this test, whose vendors were confident of their ability to protect against these threats in the test.
Differences between the MITRE ATT&CK® Test and our ATP Test
Whilst our Advanced Threat Protection Test makes use of elements of the ATT&CK framework, it is a very different sort of test from the ATT&CK Test. The ATT&CK Test principally evaluates enterprise security products with investigative and response capabilities in situations where the respective vendors actively monitor the attack being performed in real time. This is sometimes also referred as “red and blue team testing”. The emphasis is very much on detecting and logging attack processes (visibility), alerting administrators, and providing data to assist with manual threat-hunting and threat-countering measures.
For the ATT&CK Test, vendors set their products to “log-only” mode, in order to find out as much as possible about the attack chain. Such tests very definitely have their uses and provide valuable data. However, protecting individual systems against infection, and thus system/data damage, is not the principle aim in such a test. We also note that ATT&CK Test does not provide a final scoring or ranking system; rather, it provides raw data for analysis.
Our ATP Test, on the other hand, aims to determine how well a security product protects the system on which it is installed in everyday use. The critical question is whether the product protects the system against the attack, whereby it is not important which protection component blocks the attack, or at which stage the attack is stopped, provided the system is not compromised. We also consider false alarms in our test.
Differences between our ATP Test and our EPR Test
Our ATP (Advanced Threat Protection) Test focusses on protection (as opposed to detection or information gathering). The stage at which the attack is blocked is not relevant, provided the system is ultimately protected. The ATP Test is run for both consumer and business products, and so is of interest to all users. Consequently, we have tried to make it easier to understand for non-expert users.
Our EPR (Endpoint Protection and Response) Test, on the other hand, does take into account which stage(s) an attack reaches before being detected and blocked. It also looks at any responses made, and considers total cost of ownership. The EPR Test is only for enterprise products, and is more complex. The intended audience are IT security professionals in larger enterprises.
The following vendors participated in the Advanced Threat Protection Test. These are the vendors who were confident enough in the protection capabilities of their products against targeted attacks to take part in this public test.
Information about additional third-party engines/signatures used by some of the products: Vipre use the Bitdefender engine (in addition to their own protection features).
Most AV vendors did not participate with their respective EDR products, or disabled the EDR components of their participating products (see settings below). This may be explained by the following. The Enterprise ATP Test is an optional add-on to the Enterprise Main Test Series. We use the same product and configuration for all the tests within a series, and some EDR functions can have a negative impact on performance and false alarms.
Please note that the reached results are valid only for the products tested with their respective settings. With other settings (or products) the scores could be worse or better.
Scripts such as VBS, JS or MS Office macros can execute and install a file-less backdoor on victims’ systems and create a control channel (C2) to the attacker, who is usually in a different physical location, and maybe even in a different country. Apart from these well-known scenarios, it is possible to deliver malware using exploits, remote calls (PSexec, wmic), task scheduler, registry entries, Arduino hardware (USB RubberDucky) and WMI calls. This can be done with built-in Windows tools like PowerShell. These methods load the actual malware directly from the Internet into the target system’s memory, and continue to expand further into the local area network with native OS tools. They may even become persistent on machines in this way. This year’s test does not make use of portable executable (PE) malware. However, as the nature of advanced persistent threats continues to evolve, we may introduce one or two samples of these in the future if appropriate.
In the field of malware there are many (possibly overlapping) classification categories, and amongst other things a distinction can be made between file-based and fileless malware. Since 2017, a significant increase in fileless threats has been recorded. One reason for this is the fact that such attacks have proved very successful from the attackers’ point of view. One factor in their effectiveness is the fact that fileless threats operate only in the memory of the compromised system, making it harder for security solutions to recognize them.
Attack vectors and targets
In penetration tests, we see that certain attack vectors may not yet be well covered by security programs, and many popular AV products still provide insufficient protection. Some business security products are now making improvements in this area, and providing better protection in some scenarios. As mentioned above, we believe that consumer products also need to improve their protection against such malicious attacks; non-business users can be, and are, attacked in the same way. Anyone can be targeted, for a variety of reasons, including “doxing” (publishing confidential personal information) as an act of revenge. Attacking the home computers of businesspeople is also an obvious route into accessing their company data.
In the Advanced Threat Protection Test, we also include several different command-line stacks, CMD/PS commands, which can download malware from the network directly into RAM (staged) or base64 encoded calls. These methods completely avoid disk access, which is (usually) well guarded by security products. We sometimes use simple concealment measures, or change the method of the stager call as well. Once the malware has loaded its second stage, an http/https connection to the attacker will be established. This inside-out mechanism has the advantage of establishing a C2 channel to the attacker that is beyond the protection measures of the majority of NAT and firewall products. Once the C2 tunnel has been established, the attacker can use all known control mechanisms of the common C2 products (Meterpreter, PowerShell Empire, etc.). These include e.g. file uploads/downloads, screenshots, keylogging, Windows shell (GUI), and webcam snapshots. All the tools used are freely available. Their source code is open and created for research purposes. However, the bad guys often abuse these tools for criminal purposes.
False Positive (False Alarm) Test
A security product that blocks 100% of malicious attacks, but also blocks legitimate (non-malicious) actions, can be hugely disruptive. Consequently, we conduct a false-positives test as part of the Advanced Threat Protection Test, to check whether the tested products are able to distinguish malicious from non-malicious actions. Otherwise a security product could easily block 100% of malicious attacks that e.g. use email attachments, scripts and macros, simply by blocking such functions. For many users, this could make it impossible to carry out their normal daily tasks. Consequently, false-positive scores are taken into account in the product’s test score.
We also note that warning the user against e.g. opening harmless email attachments can lead to a “boy who cried wolf” scenario. Users who encounter a number of unnecessary warnings will sooner or later assume that all warnings are false alarms, and thus ignore a genuine warning when it comes along.
We used five different Initial Access Phases, distributed among the 15 test cases (e.g. 3 testcases came via email/spear-phishing attachment).
- Trusted Relationship: “Adversaries may breach or otherwise leverage organizations who have access to intended victims. Access through trusted third-party relationship exploits an existing connection that may not be protected or receives less scrutiny than standard mechanisms of gaining access to a network.” (Source: https://attack.mitre.org/techniques/T1199/)
- Valid accounts: “Adversaries may steal the credentials of a specific user or service account using Credential Access techniques or capture credentials earlier in their reconnaissance process through social engineering […].“ (Source: https://attack.mitre.org/techniques/T1078/)
- Replication Through Removable Media: “Adversaries may move onto systems […] by copying malware to removable media […] and renaming it to look like a legitimate file to trick users into executing it on a separate system. […]“ (Source: https://attack.mitre.org/techniques/T1091/)
- Phishing: Spearphishing Attachment: “Spearphishing attachment is […] employs the use of malware attached to an email. […]” (Source: https://attack.mitre.org/techniques/T1193/)
- Phishing: Spearphishing Link: “Spearphishing with a link […] employs the use of links to download malware contained in email […].“ (Source: https://attack.mitre.org/techniques/T1192/)
The 15 test scenarios used in this test are very briefly described below:
1) This threat is introduced via Trusted Relationship. MSHTA launches an HTML application, which executes a staged Empire PowerShell payload.
2) This threat is introduced via Trusted Relationship. A PowerShell script containing an AMSI bypass and a PowerShell Empire stager was executed.
3) This threat is introduced via Trusted Relationship. Windows Scripting Host was used to download a PowerShell payload via a integrated Empire PowerShell Stager, combined with an AMSI bypass.
4) This threat is introduced through Valid Accounts. The trusted Windows utility Microsoft Build Engine was used to proxy the execution of an Empire macro payload, which opens a command and control channel.
5) This threat is introduced through Valid Accounts. A VBScript which spawns a PowerShell process and executes an Empire payload has been used.
6) This threat is introduced through Valid Accounts. A batch file was used to execute an obfuscated PowerShell stager, download an obfuscated PoshC2
8) This threat is introduced via Removable Media (USB). MSHTA.exe executes a PowerShell stager which launches a base64-encoded PoshC2 staged PowerShell payload.
9) This threat is introduced via Removable Media (USB). A malicious Microsoft Office macro executes a PoshC2 PowerShell payload.
10) This threat is introduced via Spearphishing Attachment. VBScript downloads and executes an XSL PoshC2
11) This threat is introduced via Spearphishing Attachment. A HTML application downloads and executes an obfuscated PowerShell payload. This test case was created with Metasploit Meterpreter.
12) This threat is introduced via Spearphishing Attachment. VBScript downloads and executes an XSL payload. This test case was created with Metasploit Meterpreter.
13) This threat is introduced via Spearphishing Link. MSHTA.exe downloads and executes an obfuscated XSL payload. This test case was created with Metasploit Meterpreter.
15) This threat is introduced via Spearphishing Link. exe downloads and executes a PowerShell stager which downloads and executes an encrypted PowerShell Empire staged PowerShell payload, combined with an AMSI bypass.
False Alarm Test: Various false-alarm scenarios were used in order to see if any product is over-blocking certain actions (e.g. by blocking by policy email attachments, communication, scripts, etc.).
If during the course of the test, we were to observe products adapting their protection to our test environment, we would use countermeasures to evade these adaptations, to ensure that each product can genuinely detect the attack, as opposed to the test situation.
What is covered by the various testcases?
Our tests use a subset of the TTP (Tactics, Techniques, Procedures) listed in the MITRE ATT&CK framework. This year, the above 15 testcases cover the items shown in the table below:
For reference purposes, the full MITRE ATT&CK framework for Windows can be seen here: https://attack.mitre.org/matrices/enterprise/windows/
Below are the results for the 15 attacks used in this test:
|Threat detected, no C2 session, system protected||1 point|
|No alert shown, but no C2 session established, system protected||1 point|
|Threat not detected, C2 session established||0 points|
|Protection result invalid, as also non-malicious scripts/functions were blocked||N/A|
In our opinion, the goal of every AV/EPP/EDR system should be to detect and prevent attacks or other malware as soon as possible. In other words, if the attack is detected before, at or soon after execution, thus preventing e.g. the opening of a command and control channel, there is no need to prevent post-exploitation activities. A good burglar alarm should go off when somebody breaks into your house, not wait until they start stealing things.
A product that blocked certain legitimate functions (e.g. email attachments or scripts) in our FP test, would not be certified.
Observations on enterprise products
In this section, we report some additional information which could be of interest to readers.
Pre-execution (PRE): when the threat has not been run, and is inactive on the system.
On-execution (ON): immediately after the threat has been run.
Post-execution (POST): after the threat has been run, and its actions have been recognised.
Avast: In two cases, there was no alert, but also no stable C2-session.
Bitdefender: Many detections occurred before the threat was executed, due to heuristics for malicious scripts.
CrowdStrike: Almost all detections occurred post execution, by detecting the exploit or reverse shell.
ESET: In one case, there was no alert, but also no stable C2-session. Most of the malicious email attachments were detected before the attachments were saved to disk.
Fortinet: Detections occurred on-execution by FortiEDR. *) Fortinet’s settings for FortiEDR did not distinguish between non-malicious and malicious scripts/actions, so the results do not show how well it can block attacks without also blocking legitimate scripts/actions.
Kaspersky: About half of the attacks were blocked before the threat was executed, due to heuristics for malicious scripts, and most of the other attacks were blocked post-execution by the behaviour-blocker.
SparkCognition: Some cases were detected by AI before the threat was executed.
Vipre: Many detections occurred before the threat was executed, due to heuristics for malicious scripts.
All the tested vendors continuously implement improvements in the product, so it is to be expected that many of the missed attacks used in the test are covered by now.
Award levels reached in this ATP - Advanced Threat Protection Test
AV-Comparatives’ certification for Advanced Threat Protection is given to Approved Enterprise products which blocked at least 8 of the 15 attacks used in the Advanced Threat Protection Test, without blocking non-malicious operations. Business security programs are expected to deal with the kind of attacks used in this test, so detection of more than half of the test cases is required for certification.
About this test
The Advanced Threat Protection Test for enterprise products is an optional add-on test to the Public Enterprise Main-Test Series, i.e. only enterprise products which are in the Main-Test Series can join this add-on test. To get an overall picture of the protection capabilities of any of the tested products, readers should look at the results of the other tests in the Main-Test Series too.
As some of the attack methods used in the test make use of legitimate system programs and techniques, it would be fairly easy for a vendor to stop such attacks e.g. simply by blocking the use of these legitimate processes. However, this would result in the product concerned being marked down for false positives, in the same way that a security program would be marked down for e.g. blocking all unknown executable program files. Likewise, in this test, preventing an attack e.g. by simply blacklisting used servers, files or emails originating from a particular domain name would not be allowed as a means of preventing a targeted attack. Similarly, we do not accept an approach which does not distinguish between malicious and non-malicious processes, but requires e.g. an admin to whitelist ones that should be allowed. We note that in enterprise environments, it is possible to lock down users’ systems, e.g. to prevent the execution of PowerShell scripts or macros. The idea of a good security product is that it can distinguish between e.g. malicious and non-malicious scripts and macros, thus allowing authorised users to work efficiently whilst maintaining good security.
In the Enterprise Main-Test Series, vendors are allowed to configure the products as they see fit – as is common practice with business security products in the real world. However, precisely the same product and configuration is used for all the tests in the series. If we did not insist on this, a vendor could turn up protection settings or activate features in order to score highly in the Real-World and Malware Protection Tests, but turn them down/deactivate them for the Performance and False Positive Tests, in order to appear faster and less error-prone. In real life, users can only have one setting at once, so they should be able to see if high protection scores mean slower system performance, or lower false-positive scores mean reduced protection.
Some vendors asked for precise details of the day and time the test would be performed, so that they could monitor the attacks in real time and interact with their products when they thought it beneficial. Because the aim of the test is to measure protection capabilities, rather than analyse the attack methods, we did not provide any vendors with any advance information about when the test would be performed. In real life, attackers do not tell their victims when they are going to attack, so products must provide protection all the time. We also had requests from vendors regarding the attack methods to be used in the test. Again, because the test is about protection rather than analysis/visibility, we did not divulge specific details of the attack methods. After the test, we provide each participating vendor with sufficient data to assist them in understanding any of their missed test cases.
The test is very challenging, but at the same time it also reflects realistic scenarios. We have had positive feedback from many vendors’ technical departments. Penetration testers see the real capabilities of products in their tests every day. Our comparison test tries to create a level playing-field that allows us to fairly compare the protection capabilities of the different products against such attacks. This lets users see how well they are protected, and allows vendors, where necessary, to improve their products in the future.
Copyright and Disclaimer
This publication is Copyright © 2020 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.