AI Malware Evolves to Rewrite Code, Evade Detection

Experimental malware named PROMPTFLUX has been developed to leverage artificial intelligence, specifically Google's Gemini, to rewrite its own code in an effort to evade detection. Google's Threat Intelligence Group discovered this malware, noting that one version was designed to request code modifications every hour. This dynamic self-rewriting capability makes the malware a moving target for security software that typically relies on identifying known patterns in malicious code.

While PROMPTFLUX was still under development when discovered and had not yet successfully compromised any devices, its potential highlights a significant evolution in cyber threats. Google has since documented instances of AI being used by malware during live attacks, including an Android backdoor that uses AI to understand its environment and adapt its actions.

How PROMPTFLUX Operates

PROMPTFLUX, identified in June 2025, utilizes a component dubbed the "Thinking Robot" to interact with Gemini. This allows the malware to request new obfuscation techniques, making its code more difficult for security systems to recognize. Variations of PROMPTFLUX included instructions for Gemini to rewrite the entire source code hourly, preserving only the essential functions for continued operation. This means that even if one version of the malware is identified, it can quickly generate a new, distinct version while pursuing the same malicious goals.

Google has described PROMPTFLUX as an early example of "just-in-time" AI being integrated directly into malware. This approach allows the malware to solicit AI assistance while it is actively running, rather than relying solely on pre-written functions.

Evolving Malware and Defense Strategies

Contrary to fears that AI-powered malware could render antivirus software obsolete, security experts emphasize that detection methods are also evolving. While signature-based detection, which identifies known malware by its digital fingerprint, remains important, modern antivirus solutions employ multiple layers of defense. These include real-time monitoring, behavioral analysis, and heuristic protection, often enhanced by cloud-delivered protection and machine learning. These advanced techniques can identify new threats based on suspicious activity, even if the malware's code has changed.

Beyond PROMPTFLUX, another malware dubbed PROMPTSTEAL has been observed in real-world attacks. This malware queries a large language model to generate Windows commands that can be executed on infected systems, enabling it to gather information and exfiltrate files. This marks a significant step where AI becomes an active participant in the malware's post-infection activities.

Furthermore, PROMPTSPY, an Android backdoor identified in May 2026, features an AI module that allows it to analyze the device's screen and interact with the interface. This capability could enable malware to adapt its behavior based on what is displayed on the infected device, making it more challenging to remove.

Broader Trends in Cybercrime

Google's recent reports indicate a broader trend of attackers moving toward more automated and agentic AI workflows. This shift suggests that AI will be used to handle larger portions of cyberattack operations with less direct human involvement. One alarming example involved a financially motivated attacker using an AI coding chatbot to plan, build, and execute a mass credential-harvesting campaign in under six hours, managing vulnerability scanning and troubleshooting autonomously during the attack.

Despite these advancements, Google notes that fully autonomous exploit pipelines deployed in the wild have not yet been observed. However, the increasing automation capabilities of AI are steadily reducing the workload for attackers.

This evolving threat landscape emerges on top of an already massive volume of malware. The FBI reported nearly $21 billion in cyber-enabled crime losses in 2025, a figure that underscores the financial incentives for cybercriminals and the growing impact of technological advancements on attack efficacy.

Protecting Against Advanced Malware

To combat increasingly sophisticated malware, security experts recommend a multi-layered defense approach:

• Behavioral Antivirus Protection: Utilize antivirus software with real-time protection and behavioral detection capabilities that can identify suspicious activities even if the malware's signature is unknown.
• Enable Real-time and Cloud Protection: Ensure all security features, including real-time monitoring and cloud-delivered protection, are active and never disable them at the prompt of a website or download.
• Automatic Software Updates: Keep operating systems, browsers, and all software updated to patch known vulnerabilities that attackers often exploit.
• Caution with Commands: Never paste computer commands into websites or prompts, especially if they appear on fake error messages or CAPTCHA pages.
• Heed Security Warnings: Pay attention to browser and system security warnings, which can indicate dangerous websites or downloads.
• Secure App and Extension Sources: Download applications and browser extensions only from trusted sources, such as official app stores, and be cautious of sideloaded apps.
• Use Password Managers and MFA: Employ password managers to create strong, unique passwords and enable multi-factor authentication or passkeys for enhanced account security.
• Maintain Backups: Regularly back up important files to an external drive or cloud service to mitigate data loss from ransomware or destructive malware.
• Know Incident Response: If a device is suspected of infection, disconnect it from the internet, run a security scan, and change critical passwords from a trusted device.