Cybersecurity researchers at ESET have identified what appears to be the first AI-powered ransomware, named PromptLock. The malware employs OpenAI’s gpt-oss-20b model, accessed locally through the Ollama API, to dynamically generate malicious Lua scripts capable of file encryption and data theft.
According to ESET researchers Anton Cherepanov and Peter Strycek, the scripts run across Windows, Linux, and macOS, enabling the malware to scan systems, exfiltrate files, or encrypt data. Written in Golang, PromptLock leverages the NSA-developed SPECK 128-bit encryption algorithm and routes its operations via Ollama.
Interestingly, the ransomware’s payment request uses a Bitcoin address tied to Satoshi Nakamoto, the pseudonymous creator of Bitcoin. Early samples of both Windows and Linux variants have already been uploaded to VirusTotal.
While the ransomware is not yet fully functional—its destructive features remain unfinished—researchers warn that PromptLock represents a significant proof-of-concept. “It is our responsibility to alert the cybersecurity community about such emerging threats,” Cherepanov and Strycek noted.
The discovery comes amid rising concerns over AI-driven cybercrime. Recent studies show ransomware gangs increasingly use AI for social engineering and phishing, with attacks climbing from 20% to 25.6% in early 2025 compared to last year. Experts warn this trend could soon shift from efficiency gains to entirely new attack capabilities.
Citizen Lab researcher John Scott-Railton cautioned, “We are entering the earliest stage of threat actors exploiting local AI—and we are not prepared.”
Stay ahead of emerging cybersecurity threats. For the latest insights and updates on cloud security, follow SOC News.
News Source: ITPro.com
