https://twitter.com/ESETresearch/status/1960365364300087724
Threat researchers at ESET have identified the first known AI-powered ransomware called PromptLock, which demonstrates how cybercriminals can weaponise artificial intelligence to dynamically generate malicious scripts for cross-platform data theft and encryption operations. The experimental malware, written in Golang, leverages OpenAI’s gpt-oss:20b large language model through the Ollama API to create Lua scripts on-demand for filesystem enumeration, target file inspection, data exfiltration, and file encryption across Windows, macOS, and Linux systems, marking a significant evolution in ransomware development methodologies.
PromptLock operates by connecting to a remote server hosting the language model through a proxy tunnel, then using hard-coded prompts to instruct the AI to generate specific malicious functionality as needed during an attack. The ransomware employs the lightweight SPECK 128-bit algorithm for file encryption, an unusual choice typically reserved for RFID applications rather than traditional ransomware operations. While the malware includes provisions for data destruction capabilities, this feature has not been implemented in discovered samples, suggesting the tool remains in development rather than being actively deployed in real-world attacks.
ESET researchers believe PromptLock represents a proof-of-concept rather than an active threat, citing several indicators including the use of weak encryption, a hard-coded Bitcoin address linked to Satoshi Nakamoto, and incomplete feature implementation. Following ESET’s public disclosure, a security researcher claimed ownership of the project and indicated it had been leaked unintentionally. Despite its experimental nature, PromptLock’s emergence demonstrates the potential for AI integration in malware development, following the recent discovery of LameHug malware that uses Hugging Face API and Alibaba’s language models to generate Windows shell commands in real-time. These developments highlight how artificial intelligence is lowering barriers to cybercrime entry while providing attackers with enhanced operational flexibility, cross-platform capabilities, and improved evasion techniques.