SparkCat
- Type
- Infostealer
- Capabilities
- Credential theftExfiltrationCrypto theftDefense evasion
- Delivery
- Landing page
- Platforms
- AndroidiOS
SparkCat is a mobile information stealer targeting Android and iOS devices, primarily to obtain cryptocurrency wallet recovery phrases stored in screenshots and other gallery images. Active since at least March 2024 and publicly documented in February 2025, it is embedded as malicious SDKs or frameworks within seemingly legitimate applications, including messaging, cryptocurrency-related, and food-delivery apps. Infected applications have been distributed through Google Play, Apple’s App Store, and unofficial channels.
After obtaining photo-library permissions, SparkCat uses optical character recognition to inspect stored images. It retrieves attacker-controlled keywords, dictionaries, and filtering parameters, then selectively exfiltrates matching images to remote infrastructure. Stolen recovery phrases can enable attackers to take control of victims’ cryptocurrency wallets. Early versions used Google ML Kit for OCR on both platforms. Multilingual recognition and keyword lists supported targeting across Asia and Europe, including the UAE.
SparkCat conceals its functionality through obfuscation, encrypted configuration and communications, and a Rust-based component implementing a custom command-and-control protocol. Versions discovered in early 2026 introduced additional Android obfuscation and code virtualization, including a concealed Rust library decrypted through a custom Dalvik-like virtual machine. Updated iOS variants use Apple’s Vision framework for OCR. The newer Android variants search for Chinese, Japanese, and Korean keywords, while iOS variants search for English mnemonic phrases. No established threat group has been conclusively attributed to SparkCat.
Techniques & procedures
26 distinct techniques documented for this family, organized by ATT&CK tactic.
Initial Access
3 techniques
Initial Access
We cannot confirm with certainty whether the infection was a result of a supply chain attack or deliberate action by the developers.
The malware was distributed through unofficial sources as well as Google Play and App Store... In both the Android and iOS versions, the malicious payload was part of the app itself, not of a third-party SDK or framework. | The threat actor distributed apps containing a malicious SDK/framework... On iOS, the malicious payload is delivered as frameworks (primarily mimicking AFNetworking.framework or Alamofire.framework) or obfuscated libraries disguised as libswiftDarwin.dylib, or it can be embedded directly into the app itself.
Selon les données de télémétrie de Kaspersky, les applications infectées circulent également via des canaux de distribution tiers. Certaines pages web vont jusqu’à imiter l’interface de l’App Store lorsqu’on y accède depuis un iPhone.
Execution
2 techniques
Execution
Once a configuration has been downloaded, Spark decrypts a payload from assets and executes it in a separate thread.
In early 2026, we discovered several apps on Google Play and the App Store that contained a new version of the SparkCat crypto stealer.
Stealth
6 techniques
Stealth
It was originally obfuscated, so we statically deobfuscated it before analyzing.
It retrieves the Base64-encoded value of the ccc key... decoded and then decrypted using AES-256 in ECB mode... The decrypted value is a list of URLs...
Le cheval de Troie se déguise en application légitime pour scanner discrètement la galerie photos de l’utilisateur... Il se propage via des applications infectées en apparence anodines : messageries d’entreprise et applications de livraison de repas figurent parmi les vecteurs identifiés.
Once a configuration has been downloaded, Spark decrypts a payload from assets and executes it in a separate thread. It uses XOR with a 16-byte key for a cipher.
Обфусцированная вредоносная библиотека на Rust расшифровывалась с помощью виртуальной машины, созданной атакующими, по устройству похожей на Dalvik.
It loads different OCR models depending on the system language to recognize Latin, Korean, Chinese or Japanese characters in images.
Credential Access
2 techniques
Credential Access
The reasoning behind the SDK’s request seems sound at first: users may attach images when contacting support.
The terms all indicated that the attackers were financially motivated, specifically targeting recovery phrases also known as “mnemonics” that can be used to regain access to cryptocurrency wallets.
Discovery
5 techniques
Discovery
the handler requests access to the device’s image gallery. If the pw flag in the aforementioned object is equal to 1, the module will keep requesting access if denied.
The SDK then uploads device information to /api/e/d/u on the C2 server.
Обфусцированная вредоносная библиотека на Rust расшифровывалась с помощью виртуальной машины, созданной атакующими, по устройству похожей на Dalvik.
It loads different OCR models depending on the system language to recognize Latin, Korean, Chinese or Japanese characters in images.
MMLocationMgr : collects the current location of the device. It sent no data during our testing, so the exact purpose of this class remained unclear.
Collection
5 techniques
Collection
Some of these scanned users’ image galleries in search of crypto wallet access recovery phrases.
The reasoning behind the SDK’s request seems sound at first: users may attach images when contacting support.
the malware is flexible enough to steal not just these phrases but also other sensitive data from the gallery, such as messages or passwords that might have been captured in screenshots.
These parameters are used by processor classes that filter images by OCR-recognized words.
It would then use an OCR model to select and exfiltrate images of interest... ML Kit searched for text blocks and then broke them down into lines. If at least three lines containing a word with a minimum of three letters were found, the Trojan would send the image to the attackers’ server.
Command and Control
4 techniques
Command and Control
Spark uses POST requests to communicate with the “http” server.
The JSON is sent to the server with the help of the native libmodsvmp.so library via the unidentified protocol over TCP sockets.
When initialized, it downloads a JSON configuration file from a GitLab URL embedded in the malware body.
It encrypts data with AES-256 in CBC mode before sending... Before sending data to the server, the library generates a 32-byte key for the AES-GCM-SIV cipher.
Exfiltration
2 techniques
Exfiltration
Images that matched keywords received from the C2 were sent to the server.
Next, the image is uploaded to either Amazon’s cloud storage or to file@/api/res/send on the “rust” server.
Observables
105 indicators attributed across vendor reports, sandbox runs, and researcher write-ups. Values are masked here and available in full in Mallory.
Recent activity
18 sources tracked across advisories, community write-ups, and news. New activity surfaces here as Mallory finds it.