311/68 Thursday, August 28, 2025
Cybersecurity researchers at ESET have uncovered a new type of malware named PromptLock, identified as the first ransomware to leverage artificial intelligence (AI) as part of its attack process. This malware operates using prompt injection techniques, embedding pre-defined commands to manipulate AI models into performing malicious tasks-such as scanning local files, exfiltrating data, and encrypting information using SPECK 128-bit encryption. PromptLock is compatible with Windows, macOS, and Linux, written in Golang, and communicates with an open-source AI model (gpt-oss:20b) via the Ollama API to execute step-by-step instructions.
Researchers discovered the code on August 25 via VirusTotal, a public malware database. Based on initial analysis, ESET believes that the code currently exists as a proof-of-concept (PoC) rather than a fully deployed malware strain.
A New Threat Landscape Powered by AI
What raises concern, however, is the structure of PromptLock—highlighting an emerging trend in cyberattacks that use AI to automate scripting, analyze files for personal data, and even generate customized ransom notes based on the victim’s system contents. Researchers warn that this behavior could cause Indicators of Compromise (IoCs) to vary each time the code is executed, making detection and mitigation significantly more difficult.
ESET emphasizes that this discovery serves as a critical warning about the potential risks organizations may face if AI agents with high-level access to systems are deployed without proper security safeguards.
Source https://cyberscoop.com/prompt-lock-eset-ransomware-research-ai-powered-prompt-injection/
