ESET researchers have described the first known ransomware strain that uses generative artificial intelligence in its work. It is capable of infecting Windows, Linux, and macOS devices, demonstrating a novel cyberattack method and the current challenges of detecting it.
PromptLock is written in Golang. Its primary difference from well-known ransomware is its integration with the open LLM GPT-OSS:20b. Connection to the model is via the Ollama API, which runs on a remote server, without requiring the neural network itself to be downloaded to the infected device.
Once inside the system, PromptLock establishes a connection to the Ollama server and generates Lua scripts. These, in turn, scan the file system, analyze the drive's contents, and select data to encrypt. The scripts are created from scratch each time and are unique, making them difficult to detect.
Experts discuss key indicators that may indicate an infection:
Atypical file system access patterns;
Unexpected processes;
Bulk file encryption using SPECK;
Unusual network connections using the LLM model API server. As a countermeasure, we recommend using an EDR system focused on behavioral analysis, implementing network tunnel monitoring and blocking suspicious connections, as well as applying application whitelisting, script execution controls, maintaining offline backups, and procedures to quickly isolate infected computers.
For more information, please visit the ESET official website.
