Using capa Rules for Android Malware Detection

Mobile devices have become the go-to for daily tasks like online banking, healthcare management, and personal photo storage, making them prime targets for malicious actors seeking to exploit valuable information. Bad actors often turn to publishing and distributing malware via apps as a lucrative channel for generating illegal and/or unethical profits. 

Android takes a multi-layered approach to combating malware to help keep users safe (more later in the post), but while we continuously strengthen our defenses against malware, threat actors are persistently updating their malware to evade detection. Malware developers used to complete their entire malicious aggression using the common Android app development toolkits in Java, which is easier to detect by reversing the Java bytecode. In recent years, malware developers are increasing the use of native code to obfuscate some of the critical malware behaviors and putting their hopes on obscuration in compiled and symbol-stripped Executable and Linkable Format (ELF) files, which can be more difficult and time-consuming to reveal their true intentions.

To combat these new challenges, Android Security and Privacy Team is partnering with Mandiant FLARE to extend the open-source binary analysis tool capa to analyze native ARM ELF files targeting Android. Together, we improved existing and developed new capa rules to detect capabilities observed in Android malware, used the capa rule matches to highlight the highly suspicious code in native files, and prompted Gemini with the highlighted code behaviors for summarization to enhance our review processes for faster decisions.

In this blog post, we will describe how we leverage capa behavior-detection capabilities and state-of-art Gemini summarization by:

• Showcasing a malware sample that used various anti-analysis tricks to evade detections

• Explaining how our existing and new capa rules identify and highlighted those behaviors

• Presenting how Gemini summarizes the highlighted code for security reviews

An Illegal Gambling App Under a Music App Façade

Google Play Store ensures all published apps conform to local laws and regulations. This includes gambling apps, which are prohibited or require licenses in some areas. Developing and distributing illegal gambling apps in such areas can generate significant illicit profits, which sometimes is associated with organized crimes. To bypass Google Play Store's security-screening procedures, some gambling apps disguise themselves with harmless façades like music or casual games. These apps only reveal their gambling portals in certain geographic markets using various anti-analysis tricks. Unfortunately, dynamic analysis, such as emulation and sandbox detonation, relies on specific device configurations, and threat actors keep trying different combinations of settings to evade our detections. It's an ongoing game of cat and mouse!

In response, the Android Security and Privacy Team has evolved static analysis techniques, such as those that evaluate the behavior of a complete program and all its conditional logic. So, let's describe an app that violated Google Play Store rules and show how we can better detect and block other apps like it.

We received reports of a music app opening gambling websites for users in certain geographical areas. It used an interesting trick of hiding key behaviors in a native ELF file that has most symbols (except the exported ones) stripped and is loaded at runtime to evade detection.

When we decompiled the app into Java source code, using a tool like JEB Decompiler, we found that the app has a song-playing functionality as shown in "MainActivity" of Figure 1. This looks like benign behavior and is fully within the limits of Google Play Store policies.

However, there was a small region of initialization code that loads an ELF file as soon as the app is initialized when calling the onCreate function, as shown in com.x.y.z class of Figure 1. To fully understand the behavior of the entire app, we also had to reverse engineer the ELF file, which requires a completely different toolset.

Using a tool like Ghidra, we decompiled the ARM64 ELF file into C source code and found that this app estimates the user's geographic location using timezone information ("Code Section 1" in Figure 1). The code implements a loop that compares the user's timezone with a list of target regions ("Data Section" in Figure 1).

If the user's location matches a value in the list ("Data Section" in Figure 1), this malware:

1. Downloads an encrypted DEX file from a remote server ("Code Section 2" in Figure 1)

2. Decrypts the downloaded DEX file ("Code Section 3" in Figure 1)

3. Loads the decrypted DEX file into memory ("Code Section 4" in Figure 1)

The loaded DEX file uses further server-side cloaking techniques and finally loads a gambling website (Figure 3) to the app users. Compared to the app icon in Figure 2, it is an obvious mismatch of the app's advertised functionality.

While there are many detection technologies, such as YARA, available for identifying malware distributed in ELF files, they are less resilient to app updates or variations introduced by threat actors. Fortunately, the Android Security and Privacy Team has developed new techniques for detecting malicious Android apps by inspecting their native ELF components. For example, in the gambling app in Figure 3, there are many API calls dynamically resolved via the Java Native Interface (JNI) that interact with the Android runtime. Our detection systems recognized these cross-runtime interactions and reason about their intent. We've enumerated behaviors commonly seen in Android malware, such as making ptrace API calls, extracting device information, downloading code from remote servers to local storage, and making various cryptographic operations via JNI, turning them into capa detections we can use to identify and block Google Play Store threats.

Let's now talk a little more about how this works.

Android capa Rules

capa is a tool that detects capabilities in executable files. You run it against a compiled program, and it tells you what it thinks the program can do. For example, capa might suggest that a file is a backdoor, is capable of installing services, or relies on HTTP to communicate.

Mandiant FLARE extended capa to support BinExport2, an architecture agnostic representation of disassembled programs. This enables capa to match capabilities for additional architectures and file formats, such as those supported by Ghidra and its BinExport2 plugin, with an initial focus on ARM64 ELF files. The Android Security and Privacy Team then created new capa rules focused specifically on detecting capabilities observed in ARM64 ELF files used by various Android malware samples. These proprietary rules alongside capa's open-source rules are used to detect malware capabilities as part of internal Android malware analysis pipelines.

Referring back to the gambling app in Figure 3, the following Google proprietary rules and open-source capa rules matched the malicious functions performing cloaking techniques for further inspection.

Proprietary rules:

• Make ptrace API calls

• Extract device configuration information via JNI on Android

• Extract timezone via JNI on Android

• Encode or decode data using Base64 via JNI on Android

• Encrypt or decrypt data using Cipher API via JNI on Android

Open-source capa rules:

• Create or open file

• Write file on Linux

• Read file on Linux

• Check file permission on Linux

• Create thread

• Reference Base64 string

Instead of browsing hundreds of thousands lines of obfuscated code, our analysts were able to quickly identify the evidence of the app's wrong-doings using the function addresses matching those rules and enforced on the app.

Gemini Summaries of capa Rule Matches

Safeguarding the Android ecosystem, our Android malware analysis pipelines scan millions of ELF files in-depth every day, each one containing thousands to millions of lines in their decompiled codes. On top of the fast-evolving Gemini capabilities in malware analysis, capa rules are able to select the most interesting code for Gemini summarization, with sharpened focus on a much smaller set of the most suspicious functions.

We asked Gemini to summarize the functions matched on capa rules from the earlier gambling app with the following prompt:

Gemini responded with the following suggestions:

As seen in the Gemini output, the Android ELF behaviors are explained clearly on the functions matched on capa rules.

In this particular example, Gemini helped to:

• Accentuate the function call sequences to perform dynamic code loading, where our analysts can easily inspect the key function calls getCacheFilePath and getDexClassLoader

• Identify the timezone extraction with the additional URL parameter hint, where our analysts may try to probe the malicious payload quickly and accurately

• Describe more potential suspicious behaviors (e.g. getDexClassLoader JNI call, URL obfuscation) for further rule-writing ideas

capa rules in Android together with Gemini summarization shows great potential for further malware detection with more advanced techniques. Our analysts are closely monitoring the malware trends and techniques in the market and writing up-to-date capa rules to catch the bad actors in the wild.

Android's Multi-Layered Security Approach

Android’s ever-evolving, multi-layered security approach includes integrating advanced features and working with developers and device implementers to keep the Android platform and ecosystem safe. This includes, but is not limited to:

• Advanced built-in protections: Google Play Protect automatically scans every app on Android devices with Google Play Services, no matter the download source. This built-in protection, enabled by default, provides crucial security against malware and unwanted software. Google Play Protect scans more than 200 billion apps daily and performs real-time scanning at the code-level on novel apps to combat emerging and hidden threats, like polymorphic malware. In 2024, Google Play Protect’s real-time scanning identified more than 13 million new malicious apps from outside Google Play. 

• Google Play and developer protections from malware: To create a trusted experience for everyone on Google Play, we use our SAFE principles as a guide, incorporating multi-layered protections that are always evolving to help keep Google Play safe. These protections start with the developers themselves, who play a crucial role in building secure apps. We provide developers with best-in-class tools, best practices, and on-demand training resources for building safe, high-quality apps. Every app undergoes rigorous review and testing, with only approved apps allowed to appear in the Play Store. Before a user downloads an app from Play, users can explore its user reviews, ratings, and Data safety section on Google Play to help them make an informed decision. 

• Engagement with the security research community: Google works closely with the security community on multiple levels, including the App Defense Alliance, to advance app safety standards. Android also collaborates with Google Threat Intelligence Group (GTIG) to address emerging threats and safeguard Android users worldwide.

Equipped with the fast-evolving Gemini, our analysts are able to spend less time on those sophisticated samples, minimising the exposure for malicious apps and ensuring the safety of Android ecosystems.

Acknowledgement

Special thanks to Willi Ballenthin, Yannis Gasparis, Mike Hunhoff, and Moritz Raabe for their support.