Google’s Threat Intelligence Group recently identified a piece of experimental malware known as PROMPTFLUX. This malware could use AI to rewrite its own code, making detection more challenging. Security software typically searches for known patterns in malicious code, but if the code continuously changes, it becomes a moving target.
PROMPTFLUX was still under development when Google discovered it, and researchers had not seen it successfully compromise a device or network. Google disabled assets related to the activity. What’s noteworthy is AI’s increasing role in live malware attacks. Google documented instances where AI was used during active attacks, including an Android backdoor that utilized AI to understand and react to phone activity.
How AI Malware Operates
AI-powered malware like PROMPTFLUX shows potential for attackers to use AI for continuously changing malicious code, complicating detection efforts. Google’s report highlights malware’s evolution and emphasizes preparations for future risks.
PROMPTFLUX was a VBScript-based malware project discovered by Google in June 2025. One unique aspect was the ‘Thinking Robot,’ allowing the malware to contact Gemini for new obfuscation techniques. These methods aimed to make its code difficult for security software to identify.
Variations of PROMPTFLUX included capabilities to rewrite the entire source code, preserving necessary parts, and creating a moving target for defenders.
Rewriting Malware and Antivirus Effectiveness
AI-powered malware does not render antivirus programs useless. While appearance changes can complicate detection, antivirus tools provide real-time monitoring, behavior analysis, and heuristic protection beyond signature detection. Microsoft Defender Antivirus employs cloud-delivered protection and machine learning for threat identification.
AI malware might pose challenges for specific detection types. Security tools look beyond signatures to observe malicious behavior, ensuring multi-layered protection.
PROMPTSTEAL’s Live Operations
Unlike experimental PROMPTFLUX, PROMPTSTEAL engaged in live operations. Google identified Russian-backed group APT28 employing PROMPTSTEAL against Ukrainian targets. This malware queried a large language model during live attacks, generating Windows commands for execution.
These commands extracted data from targeted computers and returned information to attacker-controlled systems, introducing AI’s role within active malware functions.
PROMPTSPY’s Android Backdoor
Google detailed PROMPTSPY, an Android backdoor, in May 2026. It employed GeminiAutomationAgent, allowing AI to interpret phone activity. The AI module navigated the interface based on screen information and could impede its removal by applying invisible overlays on uninstall buttons.
Google has since taken action against PROMPTSPY’s actors, highlighting proactive security measures. Known versions are detected by Google Play Protect, averting widespread risk.
Technology’s Rapid Progression
Google’s latest report underscores attackers’ pursuit of automation. An example involved a financially motivated attacker exploiting cloud infrastructure, employing AI workflow over six hours to execute a credential-harvesting campaign.
Security companies face rising threats. AV-TEST registers over 450,000 new malicious programs and potentially unwanted applications daily. Although variations often comprise this count, the sheer volume accentuates the challenge.
Strengthening Defense Against AI Malware
Defending devices against AI-driven malware requires proactive steps. Here’s how:
- Antivirus Protection: Choose software with real-time analysis and behavioral detection. These capabilities identify suspicious activities regardless of initial malware appearance.
- Enable Safety Features: Utilize built-in protections like Microsoft’s Defender. Resist instructions to disable security for downloads.
- Automatic Software Updates: Regularly update systems and applications. Automatic updates counter vulnerabilities promptly.
- Command Caution: Avoid pasting computer commands when prompted by websites. This tactic often masks malware installation.
- Heed Warnings: Security alerts should not be disregarded. Investigate warnings rather than bypassing them hastily.
- App and Extension Sources: Download apps and browser extensions from reputable sources only.
- Password Management: Strong passwords deter attacks. Multifactor authentication enhances protection.
- Data Backup: Regularly back up crucial files to separate storage.
- Swift Response: Act quickly upon suspicion of infection. Disconnect devices from the internet and execute trusted security scans.
Kurt Knutsson emphasizes the evolving nature of AI malware. Security teams must adapt while adhering to proven strategies to resist potential threats. Keeping protective layers active and responding proactively will limit damages if malware infiltrates.
Kurt invites readers to reflect on security companies’ preparedness against advancing malware. Engaged readers are encouraged to share insights at CyberGuy.com.
Leave a Reply