First AI-Powered Ransomware Created Using OpenAI's gpt-oss:20b Model

A novel ransomware prototype called PromptLock, first reported publicly in late August 2025 can be described as the first known AI-powered ransomware, although the strain is a proof-of-concept rather than a widespread, active campaign.

PromptLock is written in Go and leverages a locally accessible LLM — OpenAI’s gpt-oss:20b — via the Ollama API to generate malicious Lua scripts on the fly; those scripts perform filesystem enumeration, selective file inspection, data exfiltration and file encryption, and are cross-platform (Windows, Linux, macOS). The sample embeds hard-coded prompts so the LLM synthesizes behavior at runtime, and the PoC uses SPECK 128-bit encryption and even assembles contextual ransom notes based on files found.

The security implications are significant: because the malicious logic is synthesized dynamically, Indicators of Compromise (IoCs) can vary between executions, making heuristic and signature detection much harder and increasing attacker agility; this lowers the technical barrier for would-be criminals and amplifies risks from prompt injection and other model-routing/jailbreak techniques.

‍