PROMPTSTEAL Malware

PROMPTSTEAL is an advanced information-stealing malware written in Python. Unlike traditional data miners that rely on predefined commands, this threat uses generative AI to dynamically create malicious instructions during execution. It has been linked to FROZENLAKE (APT28), a Russian state-sponsored threat actor known for carrying out cyber-espionage operations against Ukraine.

AI as a Weapon: How PROMPTSTEAL Operates

What sets PROMPTSTEAL apart is its use of a large language model (LLM) to generate its own attack commands in real-time. Instead of relying on hard-coded functions, it queries the Hugging Face API to access the Qwen2.5-Coder-32B-Instruct model, a variant of the Qwen LLM. This allows the malware to improvise commands based on execution context.

During analysis, PROMPTSTEAL was observed creating one-line Windows commands to harvest data about hardware, operating systems, services, and network configurations. It also generated commands to exfiltrate document files in various formats, including Microsoft Office, PDF, and TXT, from common directories such as Desktop, Documents, and Downloads.

Evolution and Adaptation

Researchers believe that PROMPTSTEAL marks a major step in the evolution of cyber threats. Its generative AI component allows for flexibility and unpredictability, which complicates detection and mitigation. Ongoing development has introduced changes in its obfuscation layers and Command-and-Control (C&C) mechanisms. Future versions are expected to refine these elements, enhancing stealth and persistence.

If left undetected, PROMPTSTEAL can result in serious consequences — including privacy breaches, financial damage, and identity theft.

Data-Stealing Malware Landscape

Numerous malware families focus on stealing information from infected systems. Some recent examples include:

• PS1Bot
• Scruffy
• SingleCamper
• BugSleep

These threats demonstrate that data-exfiltration features appear across all malware types. Whether broadly targeted or tailored for specific environments, they compromise system integrity and user safety. Prompt removal of detected threats remains critical.

Infection Vector and Distribution

PROMPTSTEAL has been disguised as an image-generation application. While the exact distribution vector remains uncertain, it is likely spread through multiple deceptive techniques.

Common infection methods include phishing, malicious email attachments, trojans, drive-by downloads, and rogue installers from freeware or file-sharing platforms. Attackers often rely on convincing social engineering to trick users into opening infected archives (ZIP, RAR), executables (EXE, RUN), or documents (PDF, Office, OneNote). In some cases, the malware may self-propagate via local networks or removable drives such as USB flash devices.

Preventive Security Practices

To minimize infection risk, users should:

• Obtain software exclusively from trusted and official sources, avoiding pirated or modified installers.
• Use legitimate activation and update utilities supplied by software vendors instead of third-party tools.
• Remain cautious while browsing and scrutinize unsolicited messages or attachments before opening them.
• Employ a reliable, regularly updated antivirus solution to perform frequent system scans and promptly remove detected threats.

Conclusion

PROMPTSTEAL showcases a turning point in cybercrime, where generative AI models are being weaponized to produce adaptive, intelligent malware. Its emergence underlines the growing need for proactive security measures, continuous monitoring, and awareness of novel attack trends that exploit cutting-edge technologies.