Researchers at Google have identified PROMPTFLUX, a self-rewriting malware powered by artificial intelligence, signalling a potential new era in cyber threats. Unlike traditional malware, this adaptive code evolves in real time, challenging established cybersecurity defences and marking the beginning of AI-driven offensive capabilities in digital warfare.

Written by Kasun Illankoon, Editor-in-Chief, Tech Revolt

In recent weeks, Google’s Threat Intelligence Group published a detailed report showing what it describes as the first documented cases of malware that uses artificial intelligence (AI) dynamically while running

A new phase in cyber-threat evolution

Traditionally, malware has been built with fixed code and signature-based behaviour; security software could scan for known patterns. The new malware families, however, appear to shift that model. According to Google, these now include “just-in-time” AI-driven code that can generate or modify malicious scripts during execution, rather than relying solely on pre-written routines.

One such family, identified as PROMPTFLUX, is reportedly written in VBScript and uses the Gemini large language model (LLM) API to request obfuscation and evasion code in real time. The malware saves the newly generated and obfuscated script in the Windows Startup folder for persistence, and includes modules for network and removable-drive propagation.

GTIG notes that the samples examined appear to be developmental or testing stage, with incomplete features and no confirmed widespread compromises so far.

Market and actor dynamics

Beyond the technical novelty, Google’s analysis draws attention to an expanding underground marketplace for AI-based hacking tools. These tools offer less technically-skilled attackers access to advanced capabilities such as script generation, evasion, and command-and-control design.

State-sponsored threat actors are also highlighted: groups from North Korea, Iran and China are reported to use generative AI tools both for reconnaissance and for crafting more advanced tools.

Why this matters for defence

The shift toward AI-enabled malware has several implications:

• Static signature-based detection may become less effective, since malicious behaviour can be generated or modified during execution rather than relying on fixed code.

• Behavioural monitoring, anomaly detection and tighter control over AI-model/API access may become more important for organisations.

• The fact that such capabilities are in a testing phase now suggests they could become more widespread and sophisticated over time.

The current status

At present, according to Google’s own report, the threat is not yet widespread or fully realised. The identified malware families appear to lack complete functionality and have not been seen causing large-scale incidents.

Nevertheless, security observers view the development as a clear signal of what might come: malware that adapts and evolves mid-attack, making adversaries and defenders locked in a more dynamic arms-race.