mrmoor/cyber-threat-intelligence · Datasets at Hugging Face

id int64 1 14.4k	text stringlengths 3 5.01k	entities list	relations list	Comments list
1,114	In other words, it’s slow going.	[]	[]	[]
1,115	In parallel, and to inform better hypothesis for the fingerprint generation, research is being done to determine how far single mutex analysis can take us.	[]	[]	[]
1,116	The research is ongoing but the initial results are extremely promising.	[]	[]	[]
1,117	The number of times any single mutex is used drops rapidly from the millions of samples down to thousands and from there, even further.	[]	[]	[]
1,118	Tens of thousands of the mutexes have been seen in only a single sample each.	[]	[]	[]
1,119	This results in a few hundred thousand individual mutexes available for further analysis.	[]	[]	[]
1,120	What quickly becomes apparent is that a large majority of the mutexes provide no obvious means to automatically classify them as necessarily indicative of good or bad behavior.	[]	[]	[]
1,121	And, unfortunately, the ones, which are reasonably easy for a human to identify, are so for significantly different reasons.	[]	[]	[]
1,122	For example, “autoproto_*” -- More than 20 mutexes share that preface, offering a natural fingerprint.	[]	[]	[]
1,123	“global\setup_028746_mutexitem” -- Associated solely with known malware digital signers.	[]	[]	[]
1,124	“defined_setnocandy” -- After reading mutex names for a few hours this just sticks out like a sore thumb.	[ { "id": 3460, "label": "TIME", "start_offset": 56, "end_offset": 67 } ]	[]	[]
1,125	Only the first of the examples had the mutex associated with a vast majority of malware samples.	[]	[]	[]
1,126	This implies that any fully automated association of a mutex to either benign or malware samples will itself require complex fingerprinting and confidence models.	[]	[]	[]
1,127	Full automation is always the ideal but it isn’t always necessary.	[]	[]	[]
1,128	With the appropriate tools it’s possible to enable a single researcher to continually review and categorize new mutexes.	[]	[]	[]
1,129	The initial classifications to be used are “benign”, “malware”, or “statistical”.	[]	[]	[]
1,130	Meaning that the mutex either itself indicates a benign or malware sample, or that the mutex alone is not enough to make a determination and the statistical ratio of benign to malware is the best it can offer.	[]	[]	[]
1,131	The backlog of already collected mutexes is too great for a small team of researchers to meaningfully tackle without some kind of ranking system.	[]	[]	[]
1,132	Luckily, the most objective piece of data collected about each mutex, how many samples were classified benign vs. malware, has all the information necessary to ensure that the researchers’ tackle the low hanging fruit first.	[]	[]	[]
1,133	With over hundreds of thousands of malware associations, this specific mutex is associated exclusively with the Net-worm:W32/Allaple malware family which has been around since 2006 but continues to propagate and reinvent itself through the years.	[ { "id": 46569, "label": "malware", "start_offset": 113, "end_offset": 133 }, { "id": 46570, "label": "TIME", "start_offset": 177, "end_offset": 181 } ]	[]	[]
1,134	Though the fact that the malware writer obviously named the mutex by rolling their face on the keyboard made it obvious before we'd done any further analysis that we’d found a unique identifier within the binaries.	[]	[]	[]
1,135	This malware is well documented as a powerful polymorphic worm that encrypts itself differently every time it propagates.	[]	[]	[]
1,136	The evasive nature of this malware family leads to a different file hash, import hash, and only a 20% average SSDeep hash overlap between the samples.	[ { "id": 46571, "label": "tools", "start_offset": 110, "end_offset": 116 } ]	[]	[]
1,137	But because the mutex name is set at compile time, the mutex itself offers a common thread between all of the samples we collected and analyzed.	[]	[]	[]
1,138	However, this particular mutex was associated with only a recent subset of the Allaple family.	[ { "id": 46572, "label": "malware", "start_offset": 80, "end_offset": 87 } ]	[]	[]
1,139	Unlike many other avenues of research and classification, mutex name based associations provides an almost trivial means of uniquely identifying common code blocks and thereby malware families.	[]	[]	[]
1,140	The first thing our researchers noticed was the similarity between this mutex name and the previous one.	[]	[]	[]
1,141	While programmatic analysis would have a hard time associating the two, it's obvious to a human that the same face rolling technique was used to name this mutex.	[]	[]	[]
1,142	The author simply rolled around a bit more.	[]	[]	[]
1,143	Quick follow up analysis revealed that this mutex was also associated with the Allaple malware family.	[ { "id": 3512, "label": "malware", "start_offset": 80, "end_offset": 87 } ]	[]	[]
1,144	More interestingly, it was another, non-overlapping, subset of the Allaple family.	[ { "id": 3513, "label": "malware", "start_offset": 67, "end_offset": 74 } ]	[]	[]
1,145	Several hypotheses followed directly from this observation.	[]	[]	[]
1,146	Absolute proof for a few of these hypotheses may never be realized.	[]	[]	[]
1,147	However, and lucky for us, the malware author was arrested in 2010 so several of the hypothesis can be verified.	[ { "id": 46573, "label": "TIME", "start_offset": 62, "end_offset": 66 } ]	[]	[]
1,148	The first is very likely due to the similarities present in the order of keys hit.	[]	[]	[]
1,149	Both begin with “jhd”.	[]	[]	[]
1,150	“jh” itself is more common than would be expected given that, with fingers on the home row, it requires the right index finger to move and press another key before any other key is struck.	[]	[]	[]
1,151	And “jh” is always followed by a key from the left side of the keyboard.	[]	[]	[]
1,152	These unique consistencies make it extremely improbable that two different people named these mutexes.	[]	[]	[]
1,153	The second mutex appears to be a concerted effort to make the mutex seem “more random” than the first.	[]	[]	[]
1,154	It's immediately obvious that the author didn't move his fingers/hands much while typing the first.	[]	[]	[]
1,155	It’s obvious enough that the author likely noticed it when reworking this section of code.	[]	[]	[]
1,156	It's highly improbable that one would see the second mutex and make a concerted effort to make it appear “less random”.	[]	[]	[]
1,157	And, as can be quickly verified by searching through standard virus detection logs, the mutexes did in fact appear in the hypothesized order.	[]	[]	[]
1,158	The third is likely due to the lack of overlap between mutex names.	[]	[]	[]
1,159	However, the research necessary to conclusively prove this hypothesis would be very time consuming and provide little other benefit.	[]	[]	[]
1,160	The fourth is likely due to the mutexes only appearing in a single malware family.	[]	[]	[]
1,161	If this functionality were available in some more open source setting, and was of even moderate quality, we would expect to see it used in other malware families as well.	[]	[]	[]
1,162	As this functionality has not migrated outside the Allaple family, either the quality of the code is bad (see: face rolling), or it's simply not available to other malware developers. The fifth is very likely as a change to the specific functionality these mutexes protect, with every change to any functionality, is simply not a practical method of development.	[ { "id": 46578, "label": "malware", "start_offset": 51, "end_offset": 58 } ]	[]	[]
1,163	And indeed, as with the second hypothesis, standard virus detection logs prove that each of these mutexes do span multiple variants of the worm.	[]	[]	[]
1,164	Mutex names provide a window into the entire development process and timeline for malware.	[]	[]	[]
1,165	Idiosyncrasies of the malware author become apparent, the evolution can be traced, the availability or quality of code deduced, and reuse of functionality made clear with a simple mutex.	[]	[]	[]
1,166	No other currently used method of analysis offers such a personal view into malware development.	[]	[]	[]
1,167	Mutex name analysis as a whole offers a unique look into the results of any sample classification system and the malware therein.	[]	[]	[]
1,168	While the research may never result in a fully automated decision system, it has been proven that researchers employing a hybrid approach to analysis will be able to provide critical and timely information to support the continual improvement of the classification system as a whole.	[]	[]	[]
1,169	From edge case to systematic misclassifications, mutex usage is even more generally the canary in a coal mine than was previously realized. While tedious and time consuming, combing through mutex names did come with more than a few good laughs.	[]	[]	[]
1,170	After nearly as much debate as some of the real research, we’ve whittled the list down to our favorites (For the curious, these are all malware mutexes).	[]	[]	[]
1,171	Enjoy.	[]	[]	[]
1,172	Pluguin - When penguins and wall sockets mate.	[]	[]	[]
1,173	Development environments don’t have spell check but, maybe they should.	[]	[]	[]
1,174	senna spy rock in rio 2001 virus - Subtlety.	[ { "id": 46580, "label": "malware", "start_offset": 1, "end_offset": 27 } ]	[]	[]
1,175	Overwhelming.	[]	[]	[]
1,176	There were many variations on this one, Senna’s obviously proud of his work.	[ { "id": 46581, "label": "identity", "start_offset": 40, "end_offset": 46 } ]	[]	[]
1,177	chinese-hacker-2 - We’re not sure which is worse: that this is a legitimate signature or a sad frame job.	[]	[]	[]
1,178	Either way, somebody needs their computer privileges revoked.	[]	[]	[]
1,179	mutexpolesskayaglush.svchost.comexefile\shell\open\command %1 %*@ - Putting shell code in a mutex name is right on the border of brilliant and insane.	[]	[]	[]
1,180	We’ll leave that determination to the reader.	[]	[]	[]
1,181	mr_coolface - Really, not so much, no.	[]	[]	[]
1,182	don't stop me!	[]	[]	[]
1,183	i need some money!	[]	[]	[]
1,184	- http://www.monster.com/ - don’t say we never did anything for you.	[ { "id": 46582, "label": "URL", "start_offset": 2, "end_offset": 25 } ]	[]	[]
1,185	Sign up to receive the latest news, cyber threat intelligence and research from us Please enter your email address!	[]	[]	[]
1,186	Please mark, I'm not a robot!	[]	[]	[]
1,187	By submitting this form, you agree to our Terms of Use and acknowledge our Privacy Statement.	[]	[]	[]
1,188	In-App Purchase (IAP) has become a popular way to sell services and virtual items through mobile applications.	[]	[]	[]
1,189	In the Android ecosystem, in addition to the official IAP service by Google, there are many third-party IAP Software Development Kits (SDKs) spread around the world.	[ { "id": 46584, "label": "identity", "start_offset": 69, "end_offset": 75 }, { "id": 46583, "label": "SOFTWARE", "start_offset": 7, "end_offset": 14 } ]	[]	[]
1,190	Some of these third-party SDKs provide IAP services based on existing online payment platforms.	[]	[]	[]
1,191	However, an increasingly popular method uses premium SMS.	[]	[]	[]
1,192	A primary reason for the popularity of SMS-based IAP is that it does not require Internet connectivity, just cell service.	[]	[]	[]
1,193	While this is more convenient for both users and developers, there are significant security concerns with using SMS-based IAP on Android.	[ { "id": 46585, "label": "SOFTWARE", "start_offset": 129, "end_offset": 136 } ]	[]	[]
1,194	These concerns are detailed below.	[]	[]	[]
1,195	Installing an Android app with SMS-based IAP is almost equivalent to installing an “SMS manager” (or even a “contacts manager”) app on the phone.	[ { "id": 46586, "label": "SOFTWARE", "start_offset": 15, "end_offset": 22 } ]	[]	[]
1,196	The reason is because most SMS-based IAP SDKs include comprehensive functionalities to monitor, manage, and even intercept a user’s SMS communication silently in the background, without any user interaction or knowledge.	[]	[]	[]
1,197	This is a critical point: these SMS-based IAPs are able to independently send, receive, or block any SMS to/from any cellphone on which they are installed, without the user having any idea this is occurring.	[]	[]	[]
1,198	This can include receiving instructions from a cloud server under the control of the SDK author.	[]	[]	[]
1,199	Palo Alto Networks Wildfire, through custom SMS blocking policies, recently discovered eight different SMS-based IAP SDKs with this abusive functionality.	[ { "id": 46588, "label": "SOFTWARE", "start_offset": 1, "end_offset": 28 } ]	[]	[]
1,200	Further investigation shows that these SDKs have been used in at least 25 different Android games in two third-party app markets, JoloPlay and Wanyx.	[ { "id": 46589, "label": "SOFTWARE", "start_offset": 84, "end_offset": 91 }, { "id": 46590, "label": "SOFTWARE", "start_offset": 130, "end_offset": 138 }, { "id": 46591, "label": "SOFTWARE", "start_offset": 143, "end_offset": 148 } ]	[]	[]
1,201	17 of the top 50 games in JoloPlay contain at least one of these IAP SDKs.	[ { "id": 46592, "label": "SOFTWARE", "start_offset": 26, "end_offset": 34 } ]	[]	[]
1,202	According to JoloPlay’s website, these 17 games have been downloaded and installed more than 1.7 million times.	[ { "id": 46593, "label": "SOFTWARE", "start_offset": 13, "end_offset": 22 } ]	[]	[]
1,203	Most of them are pirated or repackaged from famous Android games, such as Plants vs. Zombies 2, Jewels Maze, Fishing Joy, and Where's My Water.	[ { "id": 46594, "label": "SOFTWARE", "start_offset": 51, "end_offset": 58 }, { "id": 44936, "label": "SOFTWARE", "start_offset": 74, "end_offset": 94 }, { "id": 44937, "label": "SOFTWARE", "start_offset": 96, "end_offset": 107 }, { "id": 44938, "label": "SOFTWARE", "start_offset": 109, "end_offset": 120 }, { "id": 44939, "label": "SOFTWARE", "start_offset": 126, "end_offset": 142 } ]	[]	[]
1,204	According to Wanyx’s website, 12 of the 50 games in Wanyx’s suggested games column contain the discussed IAP SDKs.	[ { "id": 46596, "label": "identity", "start_offset": 13, "end_offset": 18 }, { "id": 46597, "label": "identity", "start_offset": 52, "end_offset": 58 } ]	[]	[]
1,205	In Wildfire, we classify apps using these IAP SDKs as Potentially Unwanted Applications (PUA).	[ { "id": 46599, "label": "SOFTWARE", "start_offset": 4, "end_offset": 12 } ]	[]	[]
1,206	The reason they are not marked as malware is because all SDKs require user interaction when making a purchase (e.g. user clicks on the agreement). However, these IAP SDKs are very dangerous to users because of what they are doing with a user’s SMS without his or her knowledge.	[]	[]	[]
1,207	Below are the details of eight SMS-based IAP SDKs discovered by Wildfire:	[ { "id": 46600, "label": "SOFTWARE", "start_offset": 65, "end_offset": 73 } ]	[]	[]
1,208	All of the identified IAP SDKs use premium SMS to provide the IAP service.	[]	[]	[]
1,209	They implement code that will send SMS to pre-defined premium numbers (Figure 1).	[]	[]	[]
1,210	This code implementation is also abstracted in an interface for developers.	[]	[]	[]
1,211	However, in addition to sending, all of these IAP SDKs also implement a BroadcastReceiver and register it for SMS_RECEIVED actions (Figure 2). In addition, Umpay, WiPay, and NgstreamPay will also register a ContentObserver to monitor all changes in an SMS inbox (Figure 3).	[ { "id": 46601, "label": "SOFTWARE", "start_offset": 158, "end_offset": 163 }, { "id": 46602, "label": "SOFTWARE", "start_offset": 165, "end_offset": 170 }, { "id": 46603, "label": "SOFTWARE", "start_offset": 176, "end_offset": 188 } ]	[]	[]
1,212	After an SMS is received, these receivers and observers will check the originating number and/or message body, and block the SMS by some policies (Figure 3 and Figure 4).	[]	[]	[]
1,213	For example, in Figure 4, the policy is defined as deleting SMS satisfying two conditions at the same time: 1. Message comes from the phone number “+8610658008” or “10658008” 2.	[]	[]	[]

1,114

In other words, it’s slow going.

[]

1,115

In parallel, and to inform better hypothesis for the fingerprint generation, research is being done to determine how far single mutex analysis can take us.

[]

1,116

The research is ongoing but the initial results are extremely promising.

[]

1,117

The number of times any single mutex is used drops rapidly from the millions of samples down to thousands and from there, even further.

[]

1,118

Tens of thousands of the mutexes have been seen in only a single sample each.

[]

1,119

This results in a few hundred thousand individual mutexes available for further analysis.

[]

1,120

What quickly becomes apparent is that a large majority of the mutexes provide no obvious means to automatically classify them as necessarily indicative of good or bad behavior.

[]

1,121

And, unfortunately, the ones, which are reasonably easy for a human to identify, are so for significantly different reasons.

[]

1,122

For example, “autoproto_*” -- More than 20 mutexes share that preface, offering a natural fingerprint.

[]

1,123

“global\setup_028746_mutexitem” -- Associated solely with known malware digital signers.

[]

1,124

“defined_setnocandy” -- After reading mutex names for a few hours this just sticks out like a sore thumb.

[ { "id": 3460, "label": "TIME", "start_offset": 56, "end_offset": 67 } ]

[]

1,125

Only the first of the examples had the mutex associated with a vast majority of malware samples.

[]

1,126

This implies that any fully automated association of a mutex to either benign or malware samples will itself require complex fingerprinting and confidence models.

[]

1,127

Full automation is always the ideal but it isn’t always necessary.

[]

1,128

With the appropriate tools it’s possible to enable a single researcher to continually review and categorize new mutexes.

[]

1,129

The initial classifications to be used are “benign”, “malware”, or “statistical”.

[]

1,130

Meaning that the mutex either itself indicates a benign or malware sample, or that the mutex alone is not enough to make a determination and the statistical ratio of benign to malware is the best it can offer.

[]

1,131

The backlog of already collected mutexes is too great for a small team of researchers to meaningfully tackle without some kind of ranking system.

[]

1,132

Luckily, the most objective piece of data collected about each mutex, how many samples were classified benign vs. malware, has all the information necessary to ensure that the researchers’ tackle the low hanging fruit first.

[]

1,133

With over hundreds of thousands of malware associations, this specific mutex is associated exclusively with the Net-worm:W32/Allaple malware family which has been around since 2006 but continues to propagate and reinvent itself through the years.

[ { "id": 46569, "label": "malware", "start_offset": 113, "end_offset": 133 }, { "id": 46570, "label": "TIME", "start_offset": 177, "end_offset": 181 } ]

[]

1,134

Though the fact that the malware writer obviously named the mutex by rolling their face on the keyboard made it obvious before we'd done any further analysis that we’d found a unique identifier within the binaries.

[]

1,135

This malware is well documented as a powerful polymorphic worm that encrypts itself differently every time it propagates.

[]

1,136

The evasive nature of this malware family leads to a different file hash, import hash, and only a 20% average SSDeep hash overlap between the samples.

[ { "id": 46571, "label": "tools", "start_offset": 110, "end_offset": 116 } ]

[]

1,137

But because the mutex name is set at compile time, the mutex itself offers a common thread between all of the samples we collected and analyzed.

[]

1,138

However, this particular mutex was associated with only a recent subset of the Allaple family.

[ { "id": 46572, "label": "malware", "start_offset": 80, "end_offset": 87 } ]

[]

1,139

Unlike many other avenues of research and classification, mutex name based associations provides an almost trivial means of uniquely identifying common code blocks and thereby malware families.

[]

1,140

The first thing our researchers noticed was the similarity between this mutex name and the previous one.

[]

1,141

While programmatic analysis would have a hard time associating the two, it's obvious to a human that the same face rolling technique was used to name this mutex.

[]

1,142

The author simply rolled around a bit more.

[]

1,143

Quick follow up analysis revealed that this mutex was also associated with the Allaple malware family.

[ { "id": 3512, "label": "malware", "start_offset": 80, "end_offset": 87 } ]

[]

1,144

More interestingly, it was another, non-overlapping, subset of the Allaple family.

[ { "id": 3513, "label": "malware", "start_offset": 67, "end_offset": 74 } ]

[]

1,145

Several hypotheses followed directly from this observation.

[]

1,146

Absolute proof for a few of these hypotheses may never be realized.

[]

1,147

However, and lucky for us, the malware author was arrested in 2010 so several of the hypothesis can be verified.

[ { "id": 46573, "label": "TIME", "start_offset": 62, "end_offset": 66 } ]

[]

1,148

The first is very likely due to the similarities present in the order of keys hit.

[]

1,149

Both begin with “jhd”.

[]

1,150

“jh” itself is more common than would be expected given that, with fingers on the home row, it requires the right index finger to move and press another key before any other key is struck.

[]

1,151

And “jh” is always followed by a key from the left side of the keyboard.

[]

1,152

These unique consistencies make it extremely improbable that two different people named these mutexes.

[]

1,153

The second mutex appears to be a concerted effort to make the mutex seem “more random” than the first.

[]

1,154

It's immediately obvious that the author didn't move his fingers/hands much while typing the first.

[]

1,155

It’s obvious enough that the author likely noticed it when reworking this section of code.

[]

1,156

It's highly improbable that one would see the second mutex and make a concerted effort to make it appear “less random”.

[]

1,157

And, as can be quickly verified by searching through standard virus detection logs, the mutexes did in fact appear in the hypothesized order.

[]

1,158

The third is likely due to the lack of overlap between mutex names.

[]

1,159

However, the research necessary to conclusively prove this hypothesis would be very time consuming and provide little other benefit.

[]

1,160

The fourth is likely due to the mutexes only appearing in a single malware family.

[]

1,161

If this functionality were available in some more open source setting, and was of even moderate quality, we would expect to see it used in other malware families as well.

[]

1,162

As this functionality has not migrated outside the Allaple family, either the quality of the code is bad (see: face rolling), or it's simply not available to other malware developers. The fifth is very likely as a change to the specific functionality these mutexes protect, with every change to any functionality, is simply not a practical method of development.

[ { "id": 46578, "label": "malware", "start_offset": 51, "end_offset": 58 } ]

[]

1,163

And indeed, as with the second hypothesis, standard virus detection logs prove that each of these mutexes do span multiple variants of the worm.

[]

1,164

Mutex names provide a window into the entire development process and timeline for malware.

[]

1,165

Idiosyncrasies of the malware author become apparent, the evolution can be traced, the availability or quality of code deduced, and reuse of functionality made clear with a simple mutex.

[]

1,166

No other currently used method of analysis offers such a personal view into malware development.

[]

1,167

Mutex name analysis as a whole offers a unique look into the results of any sample classification system and the malware therein.

[]

1,168

While the research may never result in a fully automated decision system, it has been proven that researchers employing a hybrid approach to analysis will be able to provide critical and timely information to support the continual improvement of the classification system as a whole.

[]

1,169

From edge case to systematic misclassifications, mutex usage is even more generally the canary in a coal mine than was previously realized. While tedious and time consuming, combing through mutex names did come with more than a few good laughs.

[]

1,170

After nearly as much debate as some of the real research, we’ve whittled the list down to our favorites (For the curious, these are all malware mutexes).

[]

1,171

Enjoy.

[]

1,172

Pluguin - When penguins and wall sockets mate.

[]

1,173

Development environments don’t have spell check but, maybe they should.

[]

1,174

senna spy rock in rio 2001 virus - Subtlety.

[ { "id": 46580, "label": "malware", "start_offset": 1, "end_offset": 27 } ]

[]

1,175

Overwhelming.

[]

1,176

There were many variations on this one, Senna’s obviously proud of his work.

[ { "id": 46581, "label": "identity", "start_offset": 40, "end_offset": 46 } ]

[]

1,177

chinese-hacker-2 - We’re not sure which is worse: that this is a legitimate signature or a sad frame job.

[]

1,178

Either way, somebody needs their computer privileges revoked.

[]

1,179

mutexpolesskayaglush*.*svchost.comexefile\shell\open\command %1 %*@ - Putting shell code in a mutex name is right on the border of brilliant and insane.

[]

1,180

We’ll leave that determination to the reader.

[]

1,181

mr_coolface - Really, not so much, no.

[]

1,182

don't stop me!

[]

1,183

i need some money!

[]

1,184

- http://www.monster.com/ - don’t say we never did anything for you.

[ { "id": 46582, "label": "URL", "start_offset": 2, "end_offset": 25 } ]

[]

1,185

Sign up to receive the latest news, cyber threat intelligence and research from us Please enter your email address!

[]

1,186

Please mark, I'm not a robot!

[]

1,187

By submitting this form, you agree to our Terms of Use and acknowledge our Privacy Statement.

[]

1,188

In-App Purchase (IAP) has become a popular way to sell services and virtual items through mobile applications.

[]

1,189

In the Android ecosystem, in addition to the official IAP service by Google, there are many third-party IAP Software Development Kits (SDKs) spread around the world.

[ { "id": 46584, "label": "identity", "start_offset": 69, "end_offset": 75 }, { "id": 46583, "label": "SOFTWARE", "start_offset": 7, "end_offset": 14 } ]

[]

1,190

Some of these third-party SDKs provide IAP services based on existing online payment platforms.

[]

1,191

However, an increasingly popular method uses premium SMS.

[]

1,192

A primary reason for the popularity of SMS-based IAP is that it does not require Internet connectivity, just cell service.

[]

1,193

While this is more convenient for both users and developers, there are significant security concerns with using SMS-based IAP on Android.

[ { "id": 46585, "label": "SOFTWARE", "start_offset": 129, "end_offset": 136 } ]

[]

1,194

These concerns are detailed below.

[]

1,195

Installing an Android app with SMS-based IAP is almost equivalent to installing an “SMS manager” (or even a “contacts manager”) app on the phone.

[ { "id": 46586, "label": "SOFTWARE", "start_offset": 15, "end_offset": 22 } ]

[]

1,196

The reason is because most SMS-based IAP SDKs include comprehensive functionalities to monitor, manage, and even intercept a user’s SMS communication silently in the background, without any user interaction or knowledge.

[]

1,197

This is a critical point: these SMS-based IAPs are able to independently send, receive, or block any SMS to/from any cellphone on which they are installed, without the user having any idea this is occurring.

[]

1,198

This can include receiving instructions from a cloud server under the control of the SDK author.

[]

1,199

Palo Alto Networks Wildfire, through custom SMS blocking policies, recently discovered eight different SMS-based IAP SDKs with this abusive functionality.

[ { "id": 46588, "label": "SOFTWARE", "start_offset": 1, "end_offset": 28 } ]

[]

1,200

Further investigation shows that these SDKs have been used in at least 25 different Android games in two third-party app markets, JoloPlay and Wanyx.

[ { "id": 46589, "label": "SOFTWARE", "start_offset": 84, "end_offset": 91 }, { "id": 46590, "label": "SOFTWARE", "start_offset": 130, "end_offset": 138 }, { "id": 46591, "label": "SOFTWARE", "start_offset": 143, "end_offset": 148 } ]

[]

1,201

17 of the top 50 games in JoloPlay contain at least one of these IAP SDKs.

[ { "id": 46592, "label": "SOFTWARE", "start_offset": 26, "end_offset": 34 } ]

[]

1,202

According to JoloPlay’s website, these 17 games have been downloaded and installed more than 1.7 million times.

[ { "id": 46593, "label": "SOFTWARE", "start_offset": 13, "end_offset": 22 } ]

[]

1,203

Most of them are pirated or repackaged from famous Android games, such as Plants vs. Zombies 2, Jewels Maze, Fishing Joy, and Where's My Water.

[ { "id": 46594, "label": "SOFTWARE", "start_offset": 51, "end_offset": 58 }, { "id": 44936, "label": "SOFTWARE", "start_offset": 74, "end_offset": 94 }, { "id": 44937, "label": "SOFTWARE", "start_offset": 96, "end_offset": 107 }, { "id": 44938, "label": "SOFTWARE", "start_offset": 109, "end_offset": 120 }, { "id": 44939, "label": "SOFTWARE", "start_offset": 126, "end_offset": 142 } ]

[]

1,204

According to Wanyx’s website, 12 of the 50 games in Wanyx’s suggested games column contain the discussed IAP SDKs.

[ { "id": 46596, "label": "identity", "start_offset": 13, "end_offset": 18 }, { "id": 46597, "label": "identity", "start_offset": 52, "end_offset": 58 } ]

[]

1,205

In Wildfire, we classify apps using these IAP SDKs as Potentially Unwanted Applications (PUA).

[ { "id": 46599, "label": "SOFTWARE", "start_offset": 4, "end_offset": 12 } ]

[]

1,206

The reason they are not marked as malware is because all SDKs require user interaction when making a purchase (e.g. user clicks on the agreement). However, these IAP SDKs are very dangerous to users because of what they are doing with a user’s SMS without his or her knowledge.

[]

1,207

Below are the details of eight SMS-based IAP SDKs discovered by Wildfire:

[ { "id": 46600, "label": "SOFTWARE", "start_offset": 65, "end_offset": 73 } ]

[]

1,208

All of the identified IAP SDKs use premium SMS to provide the IAP service.

[]

1,209

They implement code that will send SMS to pre-defined premium numbers (Figure 1).

[]

1,210

This code implementation is also abstracted in an interface for developers.

[]

1,211

However, in addition to sending, all of these IAP SDKs also implement a BroadcastReceiver and register it for SMS_RECEIVED actions (Figure 2). In addition, Umpay, WiPay, and NgstreamPay will also register a ContentObserver to monitor all changes in an SMS inbox (Figure 3).

[ { "id": 46601, "label": "SOFTWARE", "start_offset": 158, "end_offset": 163 }, { "id": 46602, "label": "SOFTWARE", "start_offset": 165, "end_offset": 170 }, { "id": 46603, "label": "SOFTWARE", "start_offset": 176, "end_offset": 188 } ]

[]

1,212

After an SMS is received, these receivers and observers will check the originating number and/or message body, and block the SMS by some policies (Figure 3 and Figure 4).

[]

1,213

For example, in Figure 4, the policy is defined as deleting SMS satisfying two conditions at the same time: 1. Message comes from the phone number “+8610658008” or “10658008” 2.

[]