Analytics Story: PromptLock

Date: 2025-09-09 ID: e86c8a7b-28f3-4aca-b6fa-50f4e8af2d2e Author: Teoderick Contreras, Splunk Product: Splunk Enterprise Security

Description

PromptLock is a proof-of-concept ransomware identified by ESET in August 2025, marking the first known instance of malware utilizing generative artificial intelligence (GenAI) for attack execution. Unlike traditional ransomware, PromptLock employs a locally hosted AI language model, specifically OpenAI's gpt-oss:20b, accessed via the Ollama API, to dynamically generate malicious Lua scripts in real time. These scripts are compatible across multiple platforms, including Windows, Linux, and macOS. During an infection, PromptLock autonomously determines which files to target for exfiltration or encryption based on predefined prompts, allowing it to adapt its behavior to the environment. The malware utilizes the SPECK 128-bit encryption algorithm and is written in Golang. While ESET considers PromptLock a proof of concept, its capabilities highlight the potential for AI to significantly enhance the sophistication and adaptability of ransomware attacks.

Why it matters

In August 2025, ESET researchers uncovered PromptLock, a proof-of-concept ransomware that represents a new frontier in cyber threats. Unlike conventional ransomware, PromptLock leverages generative artificial intelligence to autonomously create malicious scripts tailored to its environment. Using a locally hosted AI language model accessed through the Ollama API, it generates Lua scripts on the fly, enabling it to adapt dynamically to different operating systems, including Windows, macOS, and Linux. The malware can identify and target files for encryption or exfiltration based on contextual prompts, demonstrating a level of adaptability previously unseen in ransomware. Written in Golang and employing SPECK 128-bit encryption, PromptLock exemplifies how AI can enhance both the sophistication and evasiveness of malicious software. While currently a proof of concept, its discovery underscores the emerging risk of AI-driven cyberattacks and highlights the need for vigilant, forward-looking cybersecurity measures.

Detections

Name ▲▼	Technique ▲▼	Type ▲▼
Cisco NVM - Curl Execution With Insecure Flags	BITS Jobs	Anomaly
Executables Or Script Creation In Suspicious Path	Masquerading	Anomaly
Executables Or Script Creation In Temp Path	Masquerading	Anomaly
Windows Curl Upload to Remote Destination	Ingress Tool Transfer	TTP
Windows Process Execution in Temp Dir	Create or Modify System Process, Match Legitimate Resource Name or Location	Anomaly
Windows Suspicious Process File Path	Create or Modify System Process, Match Legitimate Resource Name or Location	TTP

Data Sources

Name ▲▼	Platform ▲▼	Sourcetype ▲▼	Source ▲▼
Cisco Network Visibility Module Flow Data	Network	`cisco:nvm:flowdata`	`not_applicable`
CrowdStrike ProcessRollup2	N/A	`crowdstrike:events:sensor`	`crowdstrike`
Sysmon EventID 1	Windows	`XmlWinEventLog`	`XmlWinEventLog:Microsoft-Windows-Sysmon/Operational`
Sysmon EventID 11	Windows	`XmlWinEventLog`	`XmlWinEventLog:Microsoft-Windows-Sysmon/Operational`
Windows Event Log Security 4688	Windows	`XmlWinEventLog`	`XmlWinEventLog:Security`

References

Source: GitHub | Version: 1