SE Labs

Posts tagged 'security testing'

The best security tests keep it real

Why it’s important not to try to be too clever

Latest reports now online for enterprisesmall business and home users.

Realism is important in testing, otherwise you end up with results that are theoretical and not a useful report that closely represents what is going on in the real world. One issue facing security testing that involves malware is whether or not you connect the test network to the internet.

The argument against this approach is that computer viruses can spread automatically and a test could potentially infect the real world, making life worse for computer users globally. One counter argument goes that if the tester is helping improve products then a few dozen extra infected systems on the internet is, on balance, worth it considering there are already millions out there. The benefits outweigh the downside.

Another counter argument is that viruses such as we understand them from the 90s are not the same as they are today. There are far fewer self-replicating worms and more targeted attacks that do not generally spread automatically, so the risk is lower.

Connecting to the internet brings more than a few advantages to a test, too. Firstly, the internet is where most threats reside. It would be hard to test realistically with a synthetic internet.

Secondly, for at least 10 years most endpoint security products have made connections back to management or update servers to get the latest information about current threats. So-called ‘cloud protection’ or ‘cloud updates’ would be disabled without an internet connection, effectively reducing the products’ protection abilities significantly. This then makes the test results much less accurate when running assessments.

There are cases in which turning off the internet is useful, though. Last year we ran a test to check whether or not artificial intelligence could predict future threats. We ran our Predictive Malware Response Test without an internet connection to see if a Cylance AI brain, which had been built and trained three years previously, could detect well-known threats that had come into existence since then. You can see the full report here.

But that was a special case. When assessing any security product or service for real-world, practical purposes, a live and unfiltered internet connection is probably a useful and even necessary part of the setup.

Naturally we have always used one in our testing, at one point even going as far as using consumer ADSL lines when testing home anti-malware products for extra realism. When reading security tests check that the tester has a live internet connection and allows the products to update themselves.

If you spot a detail in this report that you don’t understand, or would like to discuss, please contact us via our Twitter or Facebook accounts.

SE Labs uses current threat intelligence to make our tests as realistic as possible. To learn more about how we test, how we define ‘threat intelligence’ and how we use it to improve our tests please visit our website and follow us on Twitter.

This test report was funded by post-test consultation services provided by SE Labs to security vendors. Vendors of all products included in this report were able to request early access to results and the ability to dispute details for free. SE Labs has submitted the testing process behind this report for compliance with the AMTSO Testing Protocol Standard v1.0. To verify its compliance please check the AMTSO reference link at the bottom of page three of this report or here.

UPDATE (24th July 2019): The tests were found to be compliant with AMTSO’s Standard.

Our latest reports, for enterprisesmall business and home users are now available for free from our website. Please download them and follow us on Twitter and/or Facebook to receive updates and future reports.

How can you tell if a security test is useful or not?

How to tell if security test results are useful, misleading or just rubbish?

Latest reports now online.

In security testing circles there is a theoretical test used to illustrate how misleading some test reports can be.

For this test you need three identical chairs, packaging for three anti-virus products (in the old days products came on discs in a cardboard box) and an open window on a high floor of a building.

The methodology of this test is as follows:

  1. Tape each of the boxes to a chair. Do so carefully, such that each is fixed in exactly the same way.
  2. Throw each of the chairs out of the window, using an identical technique.
  3. Examine the chairs for damage and write a comparative report, explaining the differences found.
  4. Conclude that the best product was the one attached to the least damaged chair.

The problem with this test is obvious: the conclusions are not based on any useful reality.

The good part about this test is that the tester created a methodology and tested each product in exactly the same way.* And at least this was an ‘apples to apples’ test, in which similar products were tested in the same manner. Hopefully any tester running the chair test publishes the methodology so that readers realise what a stupidly meaningless test has been performed, but that is not a given.

Sometimes test reports come with very vague statements about, “how we tested”.

When evaluating a test report of anything, not only security products, we advise that you check how the testing was performed and to check whether or not it has been found compliant with a testing Standard, such as the Anti-Malware Testing Standards Organization’s Standard (see below).

Headline-grabbing results (e.g. Anti-virus is Dead!) catch the eye, but we need to focus on the practical realities when trying to find out how best to protect our systems from cyber threats. And that means having enough information to be able to judge a test report’s value rather than simply trusting blindly that the test was conducted correctly.

*Although some pedants might require that each chair be released from the window at exactly the same time – possible from windows far enough apart that the chairs would not entangle mid-air and skew the results in some way.

If you spot a detail in this report that you don’t understand, or would like to discuss, please contact us via our Twitter or Facebook accounts.

SE Labs uses current threat intelligence to make our tests as realistic as possible. To learn more about how we test, how we define ‘threat intelligence’ and how we use it to improve our tests please visit our website and follow us on Twitter.

These test reports were funded by post-test consultation services provided by SE Labs to security vendors. Vendors of all products included in these reports were able to request early access to results and the ability to dispute details for free. SE Labs has submitted the testing process behind this report for compliance with the AMTSO Testing Protocol Standard v1.0. To verify its compliance please check the AMTSO reference link at the bottom of page three of each report or here.

UPDATE (10th June 2019): The tests were found to be compliant with AMTSO’s Standard.

Our latest reports, for enterprise, small business and home users are now available for free from our website. Please download them and follow us on Twitter and/or Facebook to receive updates and future reports.

Enemy Unknown: Handling Customised Targeted Attacks

Detecting and preventing threats in real-time

Computer security products are designed to detect and protect against threats such as computer viruses, other malware and the actions of hackers.

A common approach is to identify existing threats and to create patterns of recognition, in much the same way as the pharmaceutical industry creates vaccinations against  known biological viruses or police issue wanted notices with photographs of known offenders.

The downside to this approach is that the virus or criminal has to be known to be harmful, most likely after someone has become sick or a crime has already been committed. It would be better to detect new infections and crimes in real-time and to stop them in action before any damage is caused.

This approach is becoming increasingly popular in the cyber security world.

Deep Instinct claims that its D-Client software is capable of detecting not only known threats but those that have not yet hit computer systems in the real world. Determining the accuracy of these claims requires a realistic test that pits the product against known threats and those typically crafted by attackers who work in a more targeted way, identifying specific potential victims and moving against them with speed and accuracy.

This test report used a range of sophisticated, high-profile threat campaigns such as those believed to have been directed against the US Presidential election in 2016, in addition to directing more targeted attacks against the victim systems using techniques seen in well-known security breaches in recent months and years.

The results show that Deep Instinct D-Client provided a wide range of detection and threat blocking capability against well-known and customised targeted attacks, without interfering with regular use of the systems upon which it was deployed. The deep learning system was  trained in August 2018, six months before the customised targeted threats were created.

Latest report now online.

Assessing next-generation protection

Malware scanning is not enough. You have to hack, too.

Latest report now online.
The amount of choice when trialling or buying endpoint security is at an all-time high. It has been 36 years since ‘anti-virus’ first appeared and, in the last five years, the number of companies innovating and selling products designed to keep Windows systems secure has exploded.
And whereas once vendors of these products generally used non-technical terms to market their wares, now computer science has come to the fore. No longer are we offered ‘anti-virus’ or ‘hacker protection’ but artificial intelligence-based detection and response solutions. The choice has never been greater, nor has the confusion among potential customers.
While marketing departments appear to have no doubt about the effectiveness of their product, the fact is that without in-depth testing no-one really knows whether or not an Endpoint Detection and Response (EDR) agent can do what it is intended.

Internal testing is necessary but inherently biased: ‘we test against what we know’. Thorough testing, including the full attack chains presented by threats, is needed to show not only detection and protection rates, but response capabilities.

EventTracker asked SE Labs to conduct an independent test of its EDR agent, running the same tests as are used against some of the world’s most established endpoint security solutions available, as well as some of the newer ones.
This report shows EventTracker’s performance in this test. The results are directly comparable with the public SE Labs Enterprise Endpoint Protection (Oct – Dec 2018) report, available here.

Can you trust security tests?

Clear, open testing is needed and now available

Latest reports now online.

A year ago we decided to put our support behind a new testing Standard proposed by the Anti-Malware Testing Standards Organization (AMTSO). The goal behind the Standard is good for everyone: if testing is conducted openly then testers such as us can receive due credit for doing a thorough job; you the reader can gain confidence in the results; and the vendors under test can understand their failings and make improvements, which then creates stronger products that we can all enjoy.

The Standard does not dictate how testers should test. There are pages of detail, but I can best summarise it like this:

Say what you are going to do, then do it. And be prepared to prove it.

(Indeed, a poor test could still comply with the AMTSO Standard, but at least you would be able to understand how the test was conducted and could then judge its worth with clear information and not marketing hype!)

We don’t think that it’s unreasonable to ask testers to make some effort to prove their results. Whether you are spending £30 on a copy of a home anti-antivirus product or several million on a new endpoint upgrade project, if you are using a report to help with your buying decision you deserve to know how the test was run, whether or not some vendors were at a disadvantage and if anyone was willing and able to double-check the results.

Since the start of the year we put our endpoint reports through the public pilot and then, once the Standard was officially adopted, through the full public process. Our last reports were judged to comply with the AMTSO Standard and we’ve submitted these latest reports for similar assessment.

At the time of writing we didn’t know if the reports from this round of testing complied. We’re pleased to report today that they did. You can confirm this by checking the AMTSO reference link at the bottom of page three of this report or here.

If you spot a detail in this report that you don’t understand, or would like to discuss, please contact us via our Twitter or Facebook accounts.

SE Labs uses current threat intelligence to make our tests as realistic as possible. To learn more about how we test, how we define ‘threat intelligence’ and how we use it to improve our tests please visit our website and follow us on Twitter.

This test report was funded by post-test consultation services provided by SE Labs to security vendors. Vendors of all products included in this report were provided with early access to results and the ability to dispute details for free. SE Labs has submitted the testing process behind this report for compliance with the AMTSO Standard v1.0.

Our latest reports, for enterprise, small business and home users are now available for free from our website. Please download them and follow us on Twitter and/or Facebook to receive updates and future reports.

How well do email security gateways protect against targeted attacks?

Email security test explores how and when services detect and stop threats.

Latest report now online.

This new email protection test shows a wide variation in the abilities of the services that we have assessed.

You might see the figures as being disappointing. Surely Microsoft Office 365 can’t be that bad? An eight per cent accuracy rating seems incredible.

Literally not credible. If it misses most threats then organisations relying on it for email security would be hacked to death (not literally).

But our results are subtler than just reflecting detection rates and it’s worth understanding exactly what we’re testing here to get the most value from the data. We’re not testing these services with live streams of real emails, in which massive percentages of messages are legitimate or basic spam. Depending on who you talk to, around 50 per cent of all email is spam. We don’t test anti-spam at all, in fact, but just the small percentage of email that comprises targeted attacks.

In other words, these results show what can happen when attackers apply themselves to specific targets. They do not reflect a “day in the life” of an average user’s email inbox.

We have also included some ‘commodity’ email threats, though – the kind of generic phishing and social engineering attacks that affect everyone. All services ought to stop every one of these. Similarly, we included some clean emails to ensure that the services were not too aggressively configured. All services ought to allow all these through to the inbox.

So when you see results that appear to be surprising, remember that we’re testing some very specific types of attacks that happen in real life, but not in vast numbers comparable to spam or more general threats.

The way that services handle threats are varied and effective to greater or lesser degrees. To best reflect how useful their responses are, we have a rating system that accounts for their different approaches. Essentially, services that keep threats as far as possible from users will win more points than those who let the message appear in or near the inbox. Conversely, those that allow the most legitimate messages through to the inbox rate higher than those which block them without the possibility of recovery from a junk folder or quarantine.

If you spot a detail in this report that you don’t understand, or would like to discuss, please contact us via our Twitter or Facebook accounts.

SE Labs uses current threat intelligence to make our tests as realistic as possible. To learn more about how we test, how we define ‘threat intelligence’ and how we use it to improve our tests please visit our website and follow us on Twitter.
Our latest reports, for enterprisesmall business and home users are now available for free from our website. Please download them and follow us on Twitter and/or Facebook to receive updates and future reports.

Latest security tests introduce attack chain scoring

When is a security breach serious, less serious or not a breach at all?

Latest reports now online.

UPDATE (29/10/2018): This set of reports are confirmed to be compliant with AMTSO Standard v1.0 by the Anti-Malware Testing Standards Organization.


Our endpoint protection tests have always included targeted attacks.

These allow us to gauge how effectively anti-malware products, in use by millions of customers, can stop hackers from breaching your systems.

We penalise products heavily for allowing partial or full breaches and, until now, that penalisation has been the same regardless of how deeply we’ve been able to penetrate into the system. Starting with this report we have updated our scoring to take varying levels of ‘success’ by us, the attackers, into account.

The new scores only apply to targeted attacks and the scoring system is listed in detail on page eight of each of the reports.

If the attackers are able to gain basic access to a target, which means they are able to run basic commands that, for example, allow them to explore the file system, then the score is -1.

The next stage is to attempt to steal a file. If successful there is a further -1 penalty.

At this stage the attackers want to take much greater control of the system. This involves increasing their account privileges – so-called privilege escalation. Success here turns a bad situation worse for the target and, if achieved, there is an additional -2 penalty.

Finally, if escalation is achieved, certain post-escalation steps are attempted, such as running a key logger or stealing passwords. A final -1 penalty is imposed if these stages are completed, making possible scores for a breach range between -1 and -5 depending on how many attack stages are possible to complete.

We have decided not to publish exact details of where in the attack chain each product stands or falls, but have provided that detailed information to the companies who produce the software tested in this report and who have asked for it.

If you spot a detail in this report that you don’t understand, or would like to discuss, please contact us via our Twitter or Facebook accounts.

SE Labs uses current threat intelligence to make our tests as realistic as possible. To learn more about how we test, how we define ‘threat intelligence’ and how we use it to improve our tests please visit our website and follow us on Twitter.
Our latest reports, for enterprisesmall business and home users are now available for free from our website. Please download them and follow us on Twitter and/or Facebook to receive updates and future reports.

Network security appliances vs. Word and PowerShell

Over the last few months we have seen a surge in attacks using apparently innocent documents that install malware covertly on victims’ systems.

Unless you are running specialist monitoring tools, or very effective security software, you probably won’t see any symptoms of the attack.

The goals of these attacks are varied. In some cases they provide remote access to hackers. In others so-called cryptocurrency mining software is installed. These programs (ab)use your systems’ processing power in an attempt to generate cryptocurrencies such as Monero. The attackers get rich off your power bill.

While there are variations in how the attacks work, the typical path to compromise involves opening the document, which could be in Microsoft Word format, after which an exploit runs a PowerShell script. This, in turn, downloads and installs the malware.

In this report we investigate how effectively some very popular network security products are at handling these and other threats.

As usual, we have also thrown in some particularly devious targeted attacks that appear to be completely legitimate applications but that provide us with remote access to unprotected targets. When we gain this access we try to hack the target in the same way a real attacker would. This gives the security products the best chance of detecting and potentially blocking the bad behaviour.

The good news is that all of these products were able to detect many (if not all) of the threats. Some were able to block most, although complete protection is not guaranteed. As always, a layered approach to protection is best. For advice on which endpoint software to choose see our Endpoint Protection test results on our website.

Latest report (PDF) now online.

Detected, blocked, quarantined, cleaned?

2018q2-1157895

What happens when your choice of security software handles an attack?

Latest reports now online.

It should be simple. You’ve clicked on the wrong link, opened a malicious email or installed something inadvisable. A threat is now attacking your PC and it’s up to your choice of anti-malware product to handle things.

But what does it actually do under the hood?

Detection is important. The product should recognise that a threat exists, even if it can’t fully handle it. At least you can receive an alert and seek help (or an alternative anti-malware program!)
Blocking threats is also very important. Ideally the protection system will prevent the malware from running. Sometimes that doesn’t happen and the malware runs. In that case one hopes that the security software would recognise that bad things are happening and stop them. This is what we call ‘neutralisation’.

Following a neutralisation your computer might not be completely clean. There could be some rogue code still on your hard disk, possibly even on your Desktop. There might also be entries in the Registry and elsewhere that will try to run this code (or code that has been deleted or quarantined).
You probably want your system to be protected by having threats blocked and, in cases where they are not, that they be removed as fast as possible and all significant traces removed. We call this happy state ‘complete remediation’.

In SE Labs tests we measure all of these outcomes, including the worst one: compromise.

If you want to know how the different products tested in this report handled threats in detail, check out the Protection Details table and graph on page 10 of our reports. We don’t show details of which products completely remediated threats and which did not when neutralising but the Protection Ratings on page eight take these into account.

If you spot a detail in this report that you don’t understand, or would like to discuss, please contact us via our Twitter or Facebook accounts.

SE Labs uses current threat intelligence to make our tests as realistic as possible. To learn more about how we test, how we define ‘threat intelligence’ and how we use it to improve our tests please visit our website and follow us on Twitter.

Our latest reports, for enterprise, small business and home users are now available for free from our website. Please download them and follow us on Twitter and/or Facebook to receive updates and future reports.

What’s the difference between SE Labs and a cyber-criminal?

nsa-jan-2018-2776476

As we prepared this network security appliance report for publication we were also getting ready to present at BT’s internal security conference Snoopcon.

We had been asked to talk about security products and how they might not do what you assume they will.

Reports like this (PDF) provide an interesting insight into how security products actually work. Marketing messages will inevitably claim world-beating levels of effectiveness, while basic tests might well support these selling points. But when you actually hack target systems through security appliances you sometimes get a very different picture.

Some vendors will support the view that testing using a full attack chain (from a malicious URL pushing an exploit, which in turn delivers a payload that finally provides us with remote access to the system) is the right way to test. Others may point out that the threats we are using don’t exactly exist in the real world of criminality because we created them in the lab and are not using them to break into systems worldwide.

We think that is a weak argument. If we can obtain access to certain popular, inexpensive tools online and create threats then these (or variants extremely close to them) are just as likely to exist in the ‘real world’ of the bad guys as in a legitimate, independent test lab. Not only that, but we don’t keep creating new threats until we break in, which is what the criminals (and penetration testers) do. We create a set and, without bias, expose all of the tested products to these threats.

But in some ways we have evolved from being anti-malware testers to being penetration testers, because we don’t just scan malware, execute scripts or visit URLs. Once we gain access to a target we perform the same tasks as a criminal would do: escalating privileges, stealing password hashes and installing keyloggers. The only difference between us and the bad guys is that we’re hacking our own systems and helping the security vendors plug the gaps.

Latest report (PDF) now online.

About

SE Labs Ltd is a private, independently-owned and run testing company that assesses security products and services. The main laboratory is located in Wimbledon, South London. It has excellent local and international travel connections. The lab is open for prearranged client visits.

Contact

SE Labs Ltd
Hill Place House
55A High Street
Wimbledon
SW19 5BA

020 3875 5000

info@selabs.uk

Press