#ESETResearch has discovered the first known AI-powered ransomware, which we named #PromptLock. The PromptLock malware uses the gpt-oss:20b model from OpenAI locally via the Ollama API to generate malicious Lua scripts on the fly, which it then executes.
PromptLock leverages Lua scripts generated from hard-coded prompts to enumerate the local filesystem, inspect target files, exfiltrate selected data, and perform encryption. These Lua scripts are cross-platform compatible, functioning on #Windows, #Linux, and #macOS.
Based on the detected user files, the malware may exfiltrate data, encrypt it, or potentially destroy it. Although the destruction functionality appears to be not yet implemented. #Bitcoin address used in the prompt appears to belong to Bitcoin creator https://en.wikipedia.org/wiki/Satoshi_Nakamoto
For its file encryption mechanism, the PromptLock ransomware utilizes the SPECK 128-bit encryption algorithm.
Although multiple indicators suggest the sample is a proof-of-concept (PoC) or work-in-progress rather than fully operational malware deployed in the wild, we believe it is our responsibility to inform the cybersecurity community about such developments.
The PromptLock ransomware is written in #Golang, and we have identified both Windows and Linux variants uploaded to VirusTotal. IoCs:
🚨 Filecoder.PromptLock.A
📄 24BF7B72F54AA5B93C6681B4F69E579A47D7C102
AD223FE2BB4563446AEE5227357BBFDC8ADA3797
BB8FB75285BCD151132A3287F2786D4D91DA58B8
F3F4C40C344695388E10CBF29DDB18EF3B61F7EF
639DBC9B365096D6347142FCAE64725BD9F73270
161CDCDB46FB8A348AEC609A86FF5823752065D2

#ESETResearch has discovered the first known AI-powered ransomware, which we named #PromptLock. The PromptLock malware uses the gpt-oss:20b model from OpenAI locally via the Ollama API to generate malicious Lua scripts on the fly, which it then executes.
PromptLock leverages Lua scripts generated from hard-coded prompts to enumerate the local filesystem, inspect target files, exfiltrate selected data, and perform encryption. These Lua scripts are cross-platform compatible, functioning on #Windows, #Linux, and #macOS.
Based on the detected user files, the malware may exfiltrate data, encrypt it, or potentially destroy it. Although the destruction functionality appears to be not yet implemented. #Bitcoin address used in the prompt appears to belong to Bitcoin creator https://en.wikipedia.org/wiki/Satoshi_Nakamoto
For its file encryption mechanism, the PromptLock ransomware utilizes the SPECK 128-bit encryption algorithm.
Although multiple indicators suggest the sample is a proof-of-concept (PoC) or work-in-progress rather than fully operational malware deployed in the wild, we believe it is our responsibility to inform the cybersecurity community about such developments.
The PromptLock ransomware is written in #Golang, and we have identified both Windows and Linux variants uploaded to VirusTotal. IoCs:
🚨 Filecoder.PromptLock.A
📄 24BF7B72F54AA5B93C6681B4F69E579A47D7C102
AD223FE2BB4563446AEE5227357BBFDC8ADA3797
BB8FB75285BCD151132A3287F2786D4D91DA58B8
F3F4C40C344695388E10CBF29DDB18EF3B61F7EF
639DBC9B365096D6347142FCAE64725BD9F73270
161CDCDB46FB8A348AEC609A86FF5823752065D2

#ESETresearch has discovered a zero-day vulnerability in WinRAR, exploited in the wild by Russia-aligned #RomCom
https://www.welivesecurity.com/en/eset-research/update-winrar-tools-now-romcom-and-others-exploiting-zero-day-vulnerability/
The vulnerability, which we assigned CVE-2025-8088, allows alternate data streams to be abused to perform path traversal. Attackers can fashion a RAR archive that, when opened, drops malicious payloads into the Windows startup directory, %TEMP%, %LOCALAPPDATA%, and others.
On July 24, we alerted the WinRAR team, which released version 7.13 just six days later. We advise all users to install the latest version as soon as possible. We would also like to thank the WinRAR team for its cooperation and quick response. https://x.com/WinRAR_RARLAB/status/1950903968923591138
Other WinRAR tools for Windows, such as the command line tools, UnRAR.dll, and the portable UnRAR source code, are also vulnerable and need to be updated.
We discovered this vulnerability being used by RomCom to deliver several different backdoors. We identified three distinct execution chains involving a SnipBot variant, RustyClaw, and the Mythic agent.
This vulnerability was also exploited by another threat actor, independently discovered by the Russian cybersecurity company BI.ZONE, who claim Paper Werewolf began using CVE-2025-8088 on July 22, just a few days after RomCom did.
https://bi.zone/expertise/blog/paper-werewolf-atakuet-rossiyu-s-ispolzovaniem-uyazvimosti-nulevogo-dnya-v-winrar/
IoCs available in our GitHub repo: https://github.com/eset/malware-ioc/tree/master/romcom

#BREAKING #ESETresearch can confirm the news of #Lumma Stealer's revival. ESET telemetry and botnet tracking show that operators are rebuilding their infrastructure, with their renewed activity reaching similar levels to those before the #disruption in May 2025.
Lumma Stealer operators are registering dozens of new domains weekly, resolving them primarily at nameservers located in RU. As seen with other threats in the past, this can complicate future attempts to disrupt its malicious operations.
In partnership with #Microsoft and other allies, ESET is tracking Lumma Stealer activity and continues to partially disrupt its operations, specifically by targeting its #Steam dead drop resolvers.
While ESET telemetry shows new Lumma Stealer builds weekly, their codebase hasn’t changed much since the takedown attempt. This suggests that the attackers had to focus on operational recovery instead of innovation.
C&C data show reduced activity around the disruption (marked in turquoise) but the number of new samples detected per day has seen only minimal changes, highlighting the continual risk.
For more info on Lumma Stealer and the disruption, read our blogpost from a few weeks back https://www.welivesecurity.com/en/eset-research/eset-takes-part-global-operation-disrupt-lumma-stealer/

#ClickFix went from virtually non-existent to the second most common attack vector blocked by #ESET, surpassed only by #phishing. This novel social engineering technique accounted for nearly 8% of all detections in H1 2025. #ESETresearch
ClickFix lures users by displaying bogus error messages followed by quick fix instructions, including copy-pasting malicious code. Running the code in the victim’s command line interpreter delivers malware such as #RATs, infostealers, and cryptominers.
Between H2 2024 and H1 2025, ESET’s detection for ClickFix, HTML/FakeCaptcha, skyrocketed by 517%. Most detections in ESET telemetry were reported from Japan (23%), Peru (6%), and Poland, Spain, and Slovakia (>5% each).
What makes #ClickFix so effective? The fake error message looks convincing; instructions are simple, yet the copied command is too technical for most users to understand. Pasting it into cmd leads to compromise with final payloads, including #DarkGate or #LummaStealer.
While #ClickFix was introduced by cybercriminals, it’s since been adopted by APT groups: Kimsuky, Lazarus; Callisto, Sednit; MuddyWater; APT36. NK-aligned actors used it to target developers, steal crypto and passwords from Metamask and #macOS Keychain.
#ClickFix uses psychological manipulation by presenting fake issues and offering quick solutions, which makes it dangerously efficient. It appears in many forms – error popups, email attachments, fake reCAPTCHAs – highlighting the need for greater vigilance online.
Read more in the #ESETThreatReport:
🔗 https://welivesecurity.com/en/eset-research/eset-threat-report-h1-2025

#ClickFix went from virtually non-existent to the second most common attack vector blocked by #ESET, surpassed only by #phishing. This novel social engineering technique accounted for nearly 8% of all detections in H1 2025. #ESETresearch
ClickFix lures users by displaying bogus error messages followed by quick fix instructions, including copy-pasting malicious code. Running the code in the victim’s command line interpreter delivers malware such as #RATs, infostealers, and cryptominers.
Between H2 2024 and H1 2025, ESET’s detection for ClickFix, HTML/FakeCaptcha, skyrocketed by 517%. Most detections in ESET telemetry were reported from Japan (23%), Peru (6%), and Poland, Spain, and Slovakia (>5% each).
What makes #ClickFix so effective? The fake error message looks convincing; instructions are simple, yet the copied command is too technical for most users to understand. Pasting it into cmd leads to compromise with final payloads, including #DarkGate or #LummaStealer.
While #ClickFix was introduced by cybercriminals, it’s since been adopted by APT groups: Kimsuky, Lazarus; Callisto, Sednit; MuddyWater; APT36. NK-aligned actors used it to target developers, steal crypto and passwords from Metamask and #macOS Keychain.
#ClickFix uses psychological manipulation by presenting fake issues and offering quick solutions, which makes it dangerously efficient. It appears in many forms – error popups, email attachments, fake reCAPTCHAs – highlighting the need for greater vigilance online.
Read more in the #ESETThreatReport:
🔗 https://welivesecurity.com/en/eset-research/eset-threat-report-h1-2025

#ESETresearch has mapped the labyrinth of #AsyncRAT forks, identifying the most prevalent versions of this open-source malware. While some variants are mere curiosities, others pose a more tenacious threat. https://www.welivesecurity.com/en/eset-research/unmasking-asyncrat-navigating-labyrinth-forks/
AsyncRAT comes with the typical RAT functionalities, including keylogging, screen capturing, and credential theft. Other threat actors have developed a multitude of variants based on its source code.
Our analysis revealed the most widely used and deployed forks of AsyncRAT, with the most prevalent among them being #DcRat.
Although DcRat holds a smaller share compared to AsyncRAT, it offers notable improvements. These include advanced evasion techniques, and the use of an open-source library for more efficient binary data serialization.
AsyncRAT forks often include prank-style plugins, such as for opening and closing the CD tray and turning off the monitor. Spoof versions dubbed SantaRAT and BoratRAT have also emerged – mostly intended as jokes.
AsyncRAT and its variants demonstrate how quickly and creatively threat actors can adapt open-source code – especially with the assistance of #LLMs. This underscores the importance of proactive detection and effective analysis of emerging threats.
IoCs available on our GitHub: https://github.com/eset/malware-ioc/tree/master/