Google Warns of PROMPTFLUX Malware That Uses Gemini API for Self-Rewriting Attacks

Google has warned of PROMPTFLUX, an experimental VBScript dropper that uses Google Gemini API for self-rewriting attacks, marking a notable shift in how AI is used in malware. According to Google Threat Intelligence Group (GTIG), PROMPTFLUX is the first confirmed case of malware that actively rewrites its own source code using a large language model.

The dropper requests obfuscation and evasion techniques from Gemini and saves the rewritten, obfuscated version to the system Startup folder to maintain persistence across reboots. It uses the gemini-1.5-flash-latest model tag to stay aligned with Gemini’s current stable release, enabling ongoing resilience against deprecation. A novel component, the Thinking Robot module, periodically queries Gemini with a hard-coded API key to obtain fresh evasion code.

Some variants replace this with a Thinging function that can instruct Gemini to regenerate the malware’s source code hourly, creating a recursive cycle of mutation. Note that while PROMPTFLUX is still experimental and has not demonstrated an ability to compromise victim networks, researchers describe its architecture as a troubling evolution in threat capabilities.