Google Uncovers PROMPTFLUX Malware That Uses Gemini AI to Rewrite Its Code Hourly

Google has uncovered an unknown threat actor using an experimental VB Script malware, dubbed PROMPTFLUX, that interacts with Gemini’s AI model API to rewrite its own code for obfuscation and evasion. According to Google Threat Intelligence Group (GTIG) the malware is written in VBScript and requests specific obfuscation techniques from Gemini to enable “just-in-time” self-modification, likely to evade static signature-based detection.

The researchers noted a “Thinking Robot” component that periodically queries Gemini 1.5 Flash or later to obtain new code, and the malware uses a hard-coded API key to reach the Gemini endpoint. In addition to regeneration, one variation uses a prompt to rewrite the entire source code every hour, with the new, obfuscated version saved to the Windows Startup folder to establish persistence, and attempts to propagate to removable drives and mapped network shares.

GTIG observed multiple PROMPTFLUX variations adopting LLM-driven code regeneration, while the broader campaign appears to be in development or testing and currently lacks proven means to compromise victim networks. The researchers also cautioned that the attacker appears financially motivated and has pursued a broad, geography- and industry-agnostic approach.