PromptLock

PromptLock is an experimental AI-enabled ransomware proof of concept, also referred to as Ransomware 3.0, that embeds a locally accessible large language model into its execution flow. It is written in Go and uses hard-coded prompts to invoke the gpt-oss-20b model through a local Ollama API, dynamically generating Lua scripts at runtime rather than relying solely on prebuilt ransomware logic. The generated scripts perform filesystem enumeration, inspect files, select data for theft, and encrypt targeted content, demonstrating how generative AI can be used to adapt malicious behavior to the victim environment during execution.

PromptLock has been described as the first known ransomware to integrate an LLM directly into active operations, but available reporting indicates it was not observed in real-world criminal deployment and was instead a research prototype closely matching an academic proof of concept developed at New York University’s Tandon School of Engineering. ESET assessed it as a proof of concept or work in progress rather than an operational campaign.

The malware is cross-platform in design. Reporting supports Windows and Linux variants, and some descriptions state the generated Lua logic was intended to be portable across Windows, Linux, and macOS. Its core demonstrated capabilities include data exfiltration and file encryption, with references to unfinished destructive or wipe-oriented logic that did not appear to be implemented in analyzed samples. PromptLock is significant primarily as an early demonstration of runtime LLM orchestration in malware, showing how publicly available local AI tooling could automate reconnaissance, target selection, and ransomware actions while potentially increasing adaptability and complicating static detection.