AI Malware Found Rewriting Its Own Code Using Gemini

Google's Threat Intelligence Group has discovered an experimental malware called PROMPTFLUX that uses generative AI 'Gemini' to rewrite its own code and evade detection. The code keeps changing in order to avoid detection. Antivirus and other security software work by identifying and blocking the characteristic patterns of known malware—so-called 'digital fingerprints'—but when code constantly changes, the malware becomes a moving target, making such pattern-based detection difficult. Some variants of PROMPTFLUX contained instructions telling Gemini to rewrite the entire source code every hour. According to researchers, however, even with its constantly changing code, the malware has not been able to fully neutralize current security tools; it has only made some types of detection more difficult.

PROMPTFLUX is a VBScript-based experimental project that Google discovered in June 2025, and it was still under development at the time of discovery. Researchers found no cases where the malware successfully compromised a victim's device or network, and Google has disabled the related assets. Still, the discovery is regarded as an example demonstrating the potential for attackers to leverage AI.

Google has since documented cases of malware using AI in the course of actual attacks. It has also identified an Android backdoor that uses AI to understand what is happening on a mobile phone and decide on its next actions on its own.