input
stringlengths
1
785
output
dict
schema
listlengths
0
0
One of the spearphishing mails we observed coming through the router network was this one, sent to an address under the gov.py (Government of Paraguay) domain.
{ "entities": { "entity": [ { "text": "One of the spearphishing mails", "start": 0, "end": 30 }, { "text": "the router network", "start": 58, "end": 76 }, { "text": "an address under the gov.py (Government of Paraguay) domain", ...
[]
There was no executable attachment in this mail, but instead a link shortened by the URL shortener service bit.ly, with the underlying link pointing to an IP address on a Dutch hosting service.
{ "entities": {} }
[]
Clicking that link from a Windows PC only yielded a redirection to the BBC homepages, and using other devices did not give more data.
{ "entities": {} }
[]
The bit.ly service does however provide information on the user creating the shortened link, and other links associated with this account.
{ "entities": {} }
[]
In this case, the user was named nicolatesla53.
{ "entities": {} }
[]
The nicolatesla53 account was created in July 2014.
{ "entities": {} }
[]
From Oct 24th to Nov 21st this user created nearly 10000 shortened links–we harvested 9990 unique ones.
{ "entities": {} }
[]
Three IP addresses were used for these links: The links themselves were on this format: As far as we were able to tell, there were three main types of actioncode: We have no sample of the actual MMS phishing messages apparently being sent, but we can see the page served when a user clicks a spammed link.
{ "entities": {} }
[]
This is just a dialogue box asking for the password presumably included in the initial message, and the next stage likely involves download of malicious content.
{ "entities": {} }
[]
We were in the middle of harvesting the servers for data on the various action codes when they all were abruptly taken offline; so our data on which mobile operators are targeted is not complete.
{ "entities": {} }
[]
We managed to get 66 of a total of 190.
{ "entities": {} }
[]
The ones we know of are shown below.
{ "entities": {} }
[]
A full breakdown of mobile operators and related links is included in the appendix.
{ "entities": {} }
[]
The composition of links created for the various mobile operators is quite interesting, as one can speculate that they represent amount of actual or planned attacks in different countries.
{ "entities": {} }
[]
With the top three operators being Vodafone, T- Mobile and Proximus (Belgacom) it seems these apparent phishing attacks are less focused on the Russian sphere than the previously discussed malware.
{ "entities": {} }
[]
This map is not complete, though.
{ "entities": {} }
[]
It represents only about 35% (66/190) of all mobile operators targeted and 66% (3152/4781) of all phishing links we managed to harvest.
{ "entities": {} }
[]
In addition, some operators like Vodafone are global actors, so the map might show an unfair intensity in their HQ locations.
{ "entities": {} }
[]
The rest of the bit.ly links used the action codes 743 or 1024.
{ "entities": {} }
[]
And now things really get interesting.
{ "entities": {} }
[]
By using mobile device HTTP User-Agents we were able to trigger downloads of malware components from some of these links.
{ "entities": {} }
[]
Accessing the link from an Android User-Agent initiated a download of an Android installer package named WhatsAppUpdate.apk.
{ "entities": { "entity": [ { "text": "Accessing the link from an Android User-Agent", "start": 0, "end": 45 }, { "text": "a download of an Android installer package named WhatsAppUpdate.apk", "start": 56, "end": 123 } ], "action": [ ...
[]
The package we analyzed was 1.2MB in size.
{ "entities": {} }
[]
The apparent main purpose of this malware is to record phone call audio.
{ "entities": { "entity": [ { "text": "this malware", "start": 29, "end": 41 }, { "text": "phone call audio", "start": 55, "end": 71 } ], "action": [ { "text": "record", "start": 48, "end": 54 } ...
[]
Recordings are stored as *.mp4 files, and uploaded to the attackers periodically.
{ "entities": { "entity": [ { "text": "Recordings", "start": 0, "end": 10 }, { "text": "*.mp4 files", "start": 25, "end": 36 }, { "text": "the attackers", "start": 54, "end": 67 }, { "text": "...
[]
The malware is able to collect a lot of other information, not all of which is actually used: Through the encrypted C & C protocol, the attackers can issue commands and binary updates to the malware.
{ "entities": { "entity": [ { "text": "The malware", "start": 0, "end": 11 }, { "text": "a lot of other information", "start": 31, "end": 57 }, { "text": "the encrypted C & C protocol", "start": 102, "end": 130 ...
[]
It uses a custom DAO/Database scheme which uses accounts belonging to the virtual community Live Journal (livejournal.com) as data stores.
{ "entities": {} }
[]
Three such accounts were found hardcoded in the package: The accounts all state that they belong to Iranian users.
{ "entities": {} }
[]
This is very likely false.
{ "entities": {} }
[]
The text in these posts starts first out in cleartext, but quickly turns into unreadable gibberish.
{ "entities": {} }
[]
The HTML source code reveals that the encoded portion is encapsulated in blog-index tags: The three accounts contain different configuration blocks pointing to C & C servers apparently located in Poland, Germany and Russia, respectively.
{ "entities": { "entity": [ { "text": "different configuration blocks", "start": 117, "end": 147 }, { "text": "C & C servers apparently located in Poland, Germany and Russia, respectively", "start": 160, "end": 236 } ], "action": [ ...
[]
Based on registration data and folder configuration we believe these are legitimate but compromised Joomla servers.
{ "entities": {} }
[]
And then an unexpected oddity shows up in the Java source: We were also able to download a similar malware sample (BrowserUpdate.apk) from one of the C & C servers.
{ "entities": {} }
[]
This sample used different online accounts for its DAO/database functionality, but is otherwise quite similar to the first.
{ "entities": {} }
[]
Using an IOS User-Agent triggered the download of a Debian installer package, WhatsAppUpdate.deb, also 1.2Mb in size.
{ "entities": { "entity": [ { "text": "Using an IOS User-Agent", "start": 0, "end": 23 }, { "text": "the download of a Debian installer package, WhatsAppUpdate.deb", "start": 34, "end": 96 } ], "action": [ { "text": "tri...
[]
This application impersonates a Cydia installer, and can only be installed on a jailbroken phone.
{ "entities": { "entity": [ { "text": "This application", "start": 0, "end": 16 }, { "text": "a Cydia installer", "start": 30, "end": 47 } ], "action": [ { "text": "impersonates", "start": 17, "end": 29 ...
[]
Once installed, it may collect These data are encrypted and uploaded to an FTP account which is taken from an encrypted configuration file named /usr/bin/cores.
{ "entities": { "entity": [ { "text": "it", "start": 16, "end": 18 }, { "text": "These data", "start": 31, "end": 41 }, { "text": "an FTP account", "start": 72, "end": 86 } ], "action": [ { ...
[]
In this particular case, the FTP account is located on a legitimate (if struggling) hosting service in the UK.
{ "entities": {} }
[]
In this case, there's another clue: The WhatsAppUpdate project seems derived from an earlier template named SkypeUpdate.
{ "entities": {} }
[]
By now, it came as no surprise when we triggered a download with a BlackBerry User-Agent.
{ "entities": {} }
[]
The initial download was a Java Applications Descriptor, a text file designed for Over-The-Air installation of Java-based applications.
{ "entities": {} }
[]
This JAD file contained the locations of the two Blackberry *.COD binaries which we then could download directly.
{ "entities": {} }
[]
The application impersonates a settings utility.
{ "entities": { "entity": [ { "text": "The application", "start": 0, "end": 15 }, { "text": "a settings utility", "start": 29, "end": 47 } ], "action": [ { "text": "impersonates", "start": 16, "end": 28 ...
[]
This collects: Collected data will be uploaded to a DynDNS domain currently hosted on a US webhosting service.
{ "entities": { "entity": [ { "text": "This", "start": 0, "end": 4 }, { "text": "Collected data", "start": 15, "end": 29 }, { "text": "a DynDNS domain currently hosted on a US webhosting service", "start": 50, "e...
[]
Since these COD files are also compiled Java code, they are possible to decompile to original source code.
{ "entities": {} }
[]
In a similar fashion to the Android version, we find interesting strings there.
{ "entities": {} }
[]
This time they are in Arabic: The earliest sample of Inception-related malware we have been able to obtain, was submitted to us in June 2014.
{ "entities": {} }
[]
However, decoy document metadata shows that it was created late May.
{ "entities": {} }
[]
The related cloud account was created just before that.
{ "entities": {} }
[]
An examination of the other documents associated with the attacks show that they have been created at a steady pace all through summer and autumn 2014 and attacks are still ongoing.
{ "entities": {} }
[]
Of interest is also the attackers' activity patterns over the 24h cycle.
{ "entities": {} }
[]
The main upload of new components to shares seems to be divided over two high–activity periods: 6:00 -10:00 UTC and 17:00-21:00 UTC.
{ "entities": {} }
[]
No uploads were seen between 23:00 and 05:00 UTC.
{ "entities": {} }
[]
It is however doubtful how indicative these timeframes are.
{ "entities": {} }
[]
To illustrate, we looked into another and more obscure timing factor: The timing of the AES InitVector random seeds.
{ "entities": {} }
[]
A random seed is the initial value passed into a pseudo-randomizer function.
{ "entities": {} }
[]
The malware uses the random output to create what is known as an InitVector-a starting point for the AES encryption/decryption function.
{ "entities": {} }
[]
The code used in some of the DLLs indicate that the attackers tend to use the C time () function to generate random seeds.
{ "entities": {} }
[]
This function returns values of granularity down to seconds.
{ "entities": {} }
[]
Thus random seeds, and ultimately the InitVectors, are functions of these quite coarse units of time.
{ "entities": {} }
[]
The encrypted files uploaded to the WebDAV shares come with their InitVectors stored at the end of the file.
{ "entities": { "entity": [ { "text": "The encrypted files", "start": 0, "end": 19 }, { "text": "the WebDAV shares", "start": 32, "end": 49 } ], "action": [ { "text": "uploaded", "start": 20, "end": 28 ...
[]
Since we know the time window to be within a few days of the upload time we were able to brute force the time values that would generate the corresponding InitVectors.
{ "entities": {} }
[]
Thus, we were able to say to the second when the file was created–and most times were identified to be in the range 1500-2200 GMT.
{ "entities": {} }
[]
Unfortunately, we had to reject these data.
{ "entities": {} }
[]
The file creation times turned out to be hours after the files themselves were uploaded to the WebDAV share.
{ "entities": { "entity": [ { "text": "the files themselves", "start": 53, "end": 73 }, { "text": "the WebDAV share", "start": 91, "end": 107 } ], "action": [ { "text": "were uploaded", "start": 74, "end"...
[]
Either the attackers' system clock is wrong or a fixed offset is added to the random seed.
{ "entities": {} }
[]
Either way, the data can't be trusted; and shows that nothing can be taken at face value.
{ "entities": {} }
[]
On at least two occasions during our surveillance of the Inception framework, the malware downloaded something unexpected and wholly different from what we have discussed until now.
{ "entities": { "entity": [ { "text": "the malware", "start": 78, "end": 89 }, { "text": "something unexpected and wholly different from what we have discussed until now", "start": 101, "end": 180 } ], "action": [ { "tex...
[]
These files were downloaded as encrypted *.bin files from the accounts carter0648 and frogs6352.
{ "entities": { "entity": [ { "text": "These files", "start": 0, "end": 11 }, { "text": "encrypted *.bin files", "start": 31, "end": 52 }, { "text": "the accounts carter0648 and frogs6352", "start": 58, "end": 95...
[]
When decrypted, these turned out to be dropper packages containing one dropper executable clearly created for the Inception framework, and one other, very different executable.
{ "entities": { "entity": [ { "text": "dropper packages", "start": 39, "end": 55 }, { "text": "one dropper executable clearly created for the Inception framework, and one other, very different executable", "start": 67, "end": 175 } ], ...
[]
This executable, (sccm.exe, md5 dd8790455109497d49c2fa2442cf16f7) is a classical Chinese APT implant.
{ "entities": {} }
[]
It is a downloader and remote shell program, designed to connect to a C & C server to interact with the attacker and/or download more malware.
{ "entities": { "entity": [ { "text": "It", "start": 0, "end": 2 }, { "text": "a downloader", "start": 6, "end": 18 }, { "text": "remote shell program", "start": 23, "end": 43 }, { "text": "a ...
[]
The C & C server in this case is ict32.msname.org.
{ "entities": { "entity": [ { "text": "The C & C server", "start": 0, "end": 16 }, { "text": "ict32.msname.org", "start": 33, "end": 49 } ], "action": [ { "text": "is", "start": 30, "end": 32 } ...
[]
When connecting to this server, sccm.exe issues the following request: This C & C domain is used by many other malwares related to sccm.exe; some of which share obvious connections to the Quarian malware family, a known APT intrusion tool.
{ "entities": { "action": [ { "text": "connecting", "start": 5, "end": 15 }, { "text": "is used", "start": 89, "end": 96 } ], "modifier": [ { "text": "to", "start": 16, "end": 18 } ], "entit...
[]
This development was unexpected for several reasons.
{ "entities": {} }
[]
First of all, it apparently breaks the strict, obfuscatory operational security built into the Inception framework.
{ "entities": {} }
[]
Inception has the capacity to perform all steps needed for scouting out and exfiltrating data without resorting to traditional hosted command & control.
{ "entities": { "entity": [ { "text": "Inception", "start": 0, "end": 9 }, { "text": "data", "start": 89, "end": 93 } ], "action": [ { "text": "scouting", "start": 59, "end": 67 }, { "...
[]
By using a well-understood APT tool and a known malicious C & C domain name, the attackers permit much clearer attribution.
{ "entities": { "action": [ { "text": "using", "start": 3, "end": 8 } ], "entity": [ { "text": "a known malicious C & C domain name", "start": 40, "end": 75 }, { "text": "the attackers", "start": 77, "end...
[]
Another factor which is out of character is the coding style.
{ "entities": {} }
[]
All Inception-related malware is written using Visual Studio 2010.
{ "entities": {} }
[]
The downloaded sccm.exe is written using Visual C++ 6; and has a PE header compile date of October 2010.
{ "entities": {} }
[]
This date can be forged, and indeed, all Inception-related malware has some level of forgery in the compile dates.
{ "entities": {} }
[]
However, the sccm.exe compile date matches the Quarian developer toolset and coding style to a better degree than the other files distributed through Inception.
{ "entities": {} }
[]
Then there is the C & C domain used.
{ "entities": { "entity": [ { "text": "the C & C domain", "start": 14, "end": 30 } ], "action": [ { "text": "used", "start": 31, "end": 35 } ] } }
[]
According to DomainTools.com the msname.org domain registration timed out September 27th 2014.
{ "entities": {} }
[]
It was left inactive and was not renewed until Nov 12th.
{ "entities": {} }
[]
This means that the attackers distributed malware that would be out of action for a long time (last distribution of sccm.exe was September 26th).
{ "entities": {} }
[]
Because of all this we consider sccm.exe as an unreliable indicator.
{ "entities": {} }
[]
It is likely to be a red herring purposefully placed on shares where the attackers have seen signs of access by security researchers.
{ "entities": {} }
[]
At one instance the attackers seem to have slipped up.
{ "entities": {} }
[]
Instead of using their scheduled task, they apparently did something manually on a WebDAV share.
{ "entities": {} }
[]
This is visible because the request came from an apparent attacker IP, but used yet another User-Agent:'' gvfs/1.12.3''.
{ "entities": {} }
[]
Gvfs is the virtual filesystem for Gnome desktop.
{ "entities": {} }
[]
The action on the account was abnormal as well; an apparent file upload: When WebDAV shares are mapped up as drives by the operating system, any action taken by the attacker follows the same pattern as on the attacker's local drive.
{ "entities": {} }
[]
In the case above, it seems the attacker attempted to edit a new document, which by default is given the name'' Untitled document'' in Gnome.
{ "entities": {} }
[]
This might indicate that the attacker's operating system language is Spanish.
{ "entities": {} }
[]
Of course, Spanish is one of the world's most widespread languages, so one can not infer much from this.
{ "entities": {} }
[]
There is even a small possibility that the phenomenon is a pure artifact; for example that a Spanish-speaking researcher connected to the same account using the same Linux-based setup as the attackers.
{ "entities": {} }
[]
This attack system shares a number of properties that are somewhat similar with the Red October campaigns detailed by Kaspersky Labs in 2013.
{ "entities": {} }
[]
For more information about this see: The'' Red October'' Campaign-An Advanced Cyber Espionage Network Targeting Diplomatic and Government Agencies However, there are also clear differences.
{ "entities": {} }
[]