As don't mind this type of articles informative and thorough, the excess use of "Windows Defender ATP" makes me think they are trying to put suggestion in the readers mind. Can we count how many times "Windows Defender ATP" is used?
I count 45 if we include variations like "Windows Defender AV", "Windows Defender Security Center", etc.
Example from article : "Windows Defender ATP alert showing how detection of the Sharpshooter technique by Windows Defender AV is surfaced in Windows Defender Security Center"
And that's not including the text embedded in the images themselves - like the image of the event log ( Event 1127 - Windows Defender ).
You can't scroll through the article without a mention of Windows Defender. I don't know if they did it purposefully or not, but I'd say it was intentional. Hard to believe it was accidental.
There's a lot of discussion of behavior based detections, but then it seems to lean heavily on AMSI, which is just an in-memory signature system.
Even the syscall monitoring, I wonder if it's really "Given a known sequence of suspicious syscalls" as opposed to "Given a known sequence of good syscalls" eg: signatures vs baselining. Or "These syscalls semantically amount to this behavior".
From what I have seen of AMSI it's not much harder than classic AV to bypass.
I appreciate the Fileless classifications - that seems like a larger value add, but I'd love to see us also better define TTP and Behavior based detections. It implies that we are observing generalized behaviors as suspicious when in reality it's almost always some sort of combination of predefined signatures.
It is not a "behavior" (in a meaningful way) when an attacker uses specific parameters on powershell or some LOLBAS executable, it's just a very specific technique, which is subject to change. The higher level "Got a process to execute and do a thing" is what the behavior is, to me, and what we should detect.
To be fair LOLBAS and Fileless are still uncommon approaches for malware, so it is not surprising that it's taking time to hammer out how we should be building out our view of the issue.
Still, ATP is a great product, not trying to be overly negative. I think Microsoft is in a really good position to start hammering out the definitions for some of these concepts. I'll give this article another read I'm sure.
AMSI, as I understand, is a way for an interpreter to cooperate with an antivirus and send code for a check. This means that if you have non-Microsoft interpreter, for example, Java or Python, then AMSI won't work (memory scanning could help though).
But what I didn't understand is why Microsoft allows users to run random JS code from Internet with full privileges. Why would anyone need to do this? This looks like a usecase made specifically for malware. If they didn't allow this, there would be no need for AMSI in the first place.
If you look at Android, there is no need for techologies like AMSI, because users cannot download and run scripts. And all applications are sandboxed and are run with minimal pivileges. Windows architecture was made for airgapped corporate environments of 80s, where there were no malware and it doesn't work in the modern era.
I wanted to write that there is no need for an antivirus on Android, but this is not true. As there are many malicious apps in Google Play and on the Internet, and users don't really understand the permission system, an antivirus that can detect malicious applications, is useful.
> what I didn't understand is why Microsoft allows users to run random JS code from Internet with full privileges
The short version is they don't. Cscript and wscript are akin to PowerShell and they aren't talking about JS from the Web.
The way it works is, say, the attacker sends the script (direct) as an email attachment and tricks the target into running it, or sends a document (indirect) that drops and runs the script, and tricks the target into opening that. They've even taken to having executables drop and run scripts to offload malicious activity into otherwise trusted processes without setting off alarms, like injection does.
In the indirect cases, cscript/wscript aren't the only options for JS. You can also:
But Microsoft is overstating the difficulty of detecting these attacks. For example obfuscation isn't a barrier to detection; in fact obfuscation can itself be detected and is a fantastic signal for blocking this stuff. General Windows system behavior, on the other hand, is chaotic[1], so in practice you're limited to writing rules for e.g. what child processes winword.exe isn't allowed to create, but even then you'll need exemptions for certain teams.
1. uTorrent for example creates child processes in a way that's practically indistinguishable from malware:
Most companies don't, for actual executables (PE files) and scripts. So for example:
1. Attackers use documents, because it isn't realistic to disallow documents over email as a matter of policy.
2. Attackers use download links (Box, OneDrive, etc.), because it isn't realistic to disallow URLs in email as a matter of policy.
3. Attackers put the naughty bits in password-protected archives, which aren't realistic for most companies to disallow over email as a matter of policy. Some will, though.
On and on. Policy is great and you should use it as much as possible, but at some point usability always wins out. Obviously there's a business opportunity there, and companies like Proofpoint and Mimecast have stepped in to fill the gap.
Pretty easy to defeat: downloaded and extracted executables are quarantined. And documents can’t ‘drop and run scripts’, they are documents. And by the way, try sending a password protected archive in the mail once. Most services just throw them away.
> downloaded and extracted executables are quarantined
The vast majority of companies won't implement a security policy that strict, even if they can distinguish email attachment downloads from other sources, which they generally can't (ADS doesn't capture it directly).
> And documents can’t ‘drop and run scripts’, they are documents
Yes they can, and do, without needing exploits, for both legitimate and illegitimate reasons. The legitimate reasons are why, again, the vast majority of companies won't implement blanket security policies here.
> And by the way, try sending a password protected archive in the mail once. Most services just throw them away.
Incorrect. You can do whatever you want with your personal email, but we're talking about enterprise security. Few companies (but more than the above two cases) will block password-protected anything as a matter of policy.
In practice, none of what you're saying will be done because of the downstream problems it creates for IT and security teams.
> The vast majority of companies won't implement a security policy that strict
Maybe such policy should be the default; by default the employee should be able to run only applications installed (or approved) by their sysadmin, and applications are sandboxed like Android's ones.
In 80s when DOS was developed, there were no easy access to the Internet, no phishing letters and viruses, so there were no risk in running different executables. But today this outdated security model simply doesn't work. You allow employees to download and run executables and they open "yearly-report.doc.exe".
Maybe desktop system developers should look at iPhones and Androids. It is difficult to dowload and run a virus on an iPhone; and it doesn't require the user to deal with configuration settings - it works out of the box. Something similar could be done for desktop systems: a pre-approved catalog of known "good" software and a simple UI (that doesn't require special skills) to whitelist any other software.
Also, I think that applications and documents should look different; today you can assign an icon of Word document to an executable; running an executable is indistinguishable from opening a document; that's wrong. Maybe applications should be located only in special application directory and run from there.
15 comments
[ 2.9 ms ] story [ 41.2 ms ] threadEdit: 38 with the image link at the bottom.
Example from article : "Windows Defender ATP alert showing how detection of the Sharpshooter technique by Windows Defender AV is surfaced in Windows Defender Security Center"
And that's not including the text embedded in the images themselves - like the image of the event log ( Event 1127 - Windows Defender ).
You can't scroll through the article without a mention of Windows Defender. I don't know if they did it purposefully or not, but I'd say it was intentional. Hard to believe it was accidental.
Even the syscall monitoring, I wonder if it's really "Given a known sequence of suspicious syscalls" as opposed to "Given a known sequence of good syscalls" eg: signatures vs baselining. Or "These syscalls semantically amount to this behavior".
From what I have seen of AMSI it's not much harder than classic AV to bypass.
I appreciate the Fileless classifications - that seems like a larger value add, but I'd love to see us also better define TTP and Behavior based detections. It implies that we are observing generalized behaviors as suspicious when in reality it's almost always some sort of combination of predefined signatures.
It is not a "behavior" (in a meaningful way) when an attacker uses specific parameters on powershell or some LOLBAS executable, it's just a very specific technique, which is subject to change. The higher level "Got a process to execute and do a thing" is what the behavior is, to me, and what we should detect.
To be fair LOLBAS and Fileless are still uncommon approaches for malware, so it is not surprising that it's taking time to hammer out how we should be building out our view of the issue.
Still, ATP is a great product, not trying to be overly negative. I think Microsoft is in a really good position to start hammering out the definitions for some of these concepts. I'll give this article another read I'm sure.
But what I didn't understand is why Microsoft allows users to run random JS code from Internet with full privileges. Why would anyone need to do this? This looks like a usecase made specifically for malware. If they didn't allow this, there would be no need for AMSI in the first place.
If you look at Android, there is no need for techologies like AMSI, because users cannot download and run scripts. And all applications are sandboxed and are run with minimal pivileges. Windows architecture was made for airgapped corporate environments of 80s, where there were no malware and it doesn't work in the modern era.
I wanted to write that there is no need for an antivirus on Android, but this is not true. As there are many malicious apps in Google Play and on the Internet, and users don't really understand the permission system, an antivirus that can detect malicious applications, is useful.
The short version is they don't. Cscript and wscript are akin to PowerShell and they aren't talking about JS from the Web.
The way it works is, say, the attacker sends the script (direct) as an email attachment and tricks the target into running it, or sends a document (indirect) that drops and runs the script, and tricks the target into opening that. They've even taken to having executables drop and run scripts to offload malicious activity into otherwise trusted processes without setting off alarms, like injection does.
In the indirect cases, cscript/wscript aren't the only options for JS. You can also:
But Microsoft is overstating the difficulty of detecting these attacks. For example obfuscation isn't a barrier to detection; in fact obfuscation can itself be detected and is a fantastic signal for blocking this stuff. General Windows system behavior, on the other hand, is chaotic[1], so in practice you're limited to writing rules for e.g. what child processes winword.exe isn't allowed to create, but even then you'll need exemptions for certain teams.1. uTorrent for example creates child processes in a way that's practically indistinguishable from malware:
1. Attackers use documents, because it isn't realistic to disallow documents over email as a matter of policy.
2. Attackers use download links (Box, OneDrive, etc.), because it isn't realistic to disallow URLs in email as a matter of policy.
3. Attackers put the naughty bits in password-protected archives, which aren't realistic for most companies to disallow over email as a matter of policy. Some will, though.
On and on. Policy is great and you should use it as much as possible, but at some point usability always wins out. Obviously there's a business opportunity there, and companies like Proofpoint and Mimecast have stepped in to fill the gap.
The vast majority of companies won't implement a security policy that strict, even if they can distinguish email attachment downloads from other sources, which they generally can't (ADS doesn't capture it directly).
> And documents can’t ‘drop and run scripts’, they are documents
Yes they can, and do, without needing exploits, for both legitimate and illegitimate reasons. The legitimate reasons are why, again, the vast majority of companies won't implement blanket security policies here.
> And by the way, try sending a password protected archive in the mail once. Most services just throw them away.
Incorrect. You can do whatever you want with your personal email, but we're talking about enterprise security. Few companies (but more than the above two cases) will block password-protected anything as a matter of policy.
In practice, none of what you're saying will be done because of the downstream problems it creates for IT and security teams.
Maybe such policy should be the default; by default the employee should be able to run only applications installed (or approved) by their sysadmin, and applications are sandboxed like Android's ones.
In 80s when DOS was developed, there were no easy access to the Internet, no phishing letters and viruses, so there were no risk in running different executables. But today this outdated security model simply doesn't work. You allow employees to download and run executables and they open "yearly-report.doc.exe".
Maybe desktop system developers should look at iPhones and Androids. It is difficult to dowload and run a virus on an iPhone; and it doesn't require the user to deal with configuration settings - it works out of the box. Something similar could be done for desktop systems: a pre-approved catalog of known "good" software and a simple UI (that doesn't require special skills) to whitelist any other software.
Also, I think that applications and documents should look different; today you can assign an icon of Word document to an executable; running an executable is indistinguishable from opening a document; that's wrong. Maybe applications should be located only in special application directory and run from there.
That’s what Windows does. You can’t not implement it.