Researchers at Google’s Threat Intelligence Group have identified a new experimental malware family, dubbed PROMPTFLUX, that uses large language models (LLMs) to modify its own code in real time to evade detection. The malware, described as a Trojan horse, interacts with Google’s Gemini AI API to dynamically rewrite, obfuscate, and generate malicious functions on demand – a technique Google calls “just-in-time” code generation. This marks a major leap toward autonomous and adaptive malware, capable of evolving faster than traditional cybersecurity defenses. While PROMPTFLUX appears to still be in its development phase and hasn’t yet been seen infecting real systems, Google warns that it could signal the rise of an underground market for illicit AI tools, lowering the barrier for less skilled hackers. The company notes that state-sponsored actors from countries including North Korea, Iran, and China are already experimenting with AI-enhanced operations. In response, Google introduced a new framework for AI security and unveiled Big Sleep, an AI agent designed to identify vulnerabilities in software, underscoring an emerging reality in cybersecurity: it’s now AI versus AI in a rapidly escalating digital arms race. Follow us 👉 @FutureTech for more! 🔌