AI Malware Rewrites Its Own Code

Google's Threat Intelligence Group has uncovered experimental malware, dubbed PROMPTFLUX, which utilizes Gemini's capabilities to rewrite its own code. This malware was designed to modify its source code hourly, a technique aimed at evading detection by security software by creating a constantly changing target. This represents an early instance of 'just-in-time' malware, leveraging AI assistance while actively running.

Further research highlighted PROMPTSTEAL, identified by Google as the first malware observed querying a large language model during active cyberattacks. This malware queries AI models to generate Windows commands for information gathering and data exfiltration. Another discovery, PROMPTSPY, features an AI module that interprets an infected Android device's interface to decide its next actions, also demonstrating sophisticated self-protection tactics.

These findings indicate a significant trend where attackers are moving towards more automated and agentic AI workflows. While fully autonomous cyberattacks are not yet observed, AI is increasingly used to automate reconnaissance and credential harvesting. For instance, a recent campaign saw an attacker plan and execute a mass credential-harvesting operation in under six hours using AI coding chatbots and agent instructions.

In response to these evolving threats, security experts recommend a multi-layered defense strategy. This includes utilizing antivirus software with behavioral detection, keeping real-time and cloud protection enabled, and ensuring all software is automatically updated. Users are strongly advised against executing unknown computer commands from websites and to heed browser security warnings.