Next-Gen Threat: Google Exposes AI-Enabled Malware That Rewrites Its Own Code with Gemini LLM

According to Google Threat Intelligence Group (GTIG), threat actors are embedding large language models into active cyber operations, moving beyond AI for productivity. The GTIG report details a new class of “just-in-time” AI-powered malware capable of rewriting its own code in real time to evade detection, with PROMPTFLUX described as a VBScript-based dropper that uses the Gemini API to rewrite its own source code hourly.

The toolchain reportedly prompts the LLM to generate and obfuscate new code, saving obfuscated versions to the Startup folder to establish persistence. In June 2025, GTIG observed APT28 (FROZENLAKE) deploying PROMPTSTEAL, a data-mining malware that queries the Qwen2.5-Coder-32B-Instruct LLM via the Hugging Face API.

The report notes that various state-backed actors, including groups from China, Iran and North Korea, have misused Gemini for reconnaissance, phishing and malware development, with examples of social-engineering and deepfake campaigns and stolen data exfiltration. Google DeepMind has since strengthened Gemini classifiers and safety guardrails, and GTIG cautions that 2025 may be remembered as a year when AI began actively influencing both sides of the cybersecurity battlefield.