Noticias

When malware learns to think

· The Next Web

Cybersecurity researchers at ESET have identified PromptSpy, marking a concerning milestone as the first Android malware to integrate generative AI directly into its operational framework. This sophisticated threat represents a fundamental shift from traditional malware approaches, demonstrating how artificial intelligence can be weaponized for malicious purposes. Unlike conventional Android trojans that rely on hardcoded coordinates and scripted sequences, PromptSpy employs Google's Gemini AI model to dynamically interpret device interfaces. The malware captures real-time screenshots of the device's current state, including buttons, labels, and layout elements, then queries Gemini for step-by-step instructions on how to interact with the interface. This AI-driven approach enables the malware to maintain persistence by keeping itself pinned in the recent-apps list, making removal significantly more challenging. The technical innovation lies in PromptSpy's ability to adapt to Android's notorious fragmentation across different manufacturers, custom UI skins, and operating system versions. Traditional malware struggles with this diversity, but by leveraging generative AI, PromptSpy can reason about unfamiliar interfaces and determine appropriate actions without human intervention. Beyond its AI capabilities, PromptSpy incorporates familiar malicious functionalities including Virtual Network Computing (VNC) modules for remote device access, lock screen data capture, screenshot recording, and overlay techniques to prevent uninstallation. However, its AI integration represents the primary differentiator and concern for security professionals. Currently, ESET characterizes PromptSpy as closer to a proof of concept than a widespread threat. The analyzed samples appear to target Argentine users specifically and have not been detected in global malware telemetry at scale. Google's Play Protect security measures reportedly flag known samples, limiting current distribution. This discovery follows ESET's previous identification of PromptLock, described as the first AI-driven ransomware for desktop systems, indicating a broader trend of AI integration in malicious software. The emergence of such threats raises critical questions about future cybersecurity challenges as AI becomes more accessible and powerful. The implications extend beyond immediate security concerns. PromptSpy demonstrates that malware developers are actively exploring AI integration, potentially leading to more adaptive and resilient threats. This development challenges traditional defense mechanisms and may require security teams to adopt AI-powered countermeasures. For the broader AI industry, PromptSpy serves as a stark reminder that technological advancement can be exploited for harmful purposes. As generative AI models become more sophisticated and widely available, the potential for malicious applications grows correspondingly. This case highlights the need for robust governance frameworks and ethical guidelines around AI deployment and access controls.

The cybersecurity landscape has reached a significant inflection point with ESET's discovery of PromptSpy, the first documented Android malware to incorporate generative artificial intelligence into its core operational logic. This breakthrough threat, unveiled by researchers in February 2026, signals a fundamental evolution in how malicious software operates and adapts to target environments.

PromptSpy distinguishes itself from conventional Android trojans through its innovative use of Google's Gemini AI model. Rather than relying on predetermined scripts or hardcoded interface coordinates, the malware captures live screenshots of device interfaces and submits them to Gemini for real-time analysis. The AI model responds with detailed instructions on how to interact with specific interface elements, enabling the malware to maintain persistence across diverse Android configurations.

This AI-driven approach addresses a longstanding challenge in Android malware development: the platform's extreme fragmentation. With countless device manufacturers, customized user interfaces, and varying operating system versions, traditional malware often fails when encountering unfamiliar layouts. PromptSpy's integration with generative AI allows it to reason about new interfaces and determine appropriate actions dynamically.

The malware's primary objective involves maintaining persistence on infected devices by manipulating the recent-apps list. Through AI-guided interactions, PromptSpy can navigate different interface designs to ensure it remains active and difficult to remove. This represents a sophisticated evolution from static automation techniques previously employed by mobile threats.

Beyond its AI capabilities, PromptSpy incorporates established malicious functionalities including Virtual Network Computing modules for remote access, comprehensive data capture mechanisms, and overlay techniques designed to prevent user-initiated removal. However, the integration of generative AI represents the most significant technical advancement and security concern.

ESET's analysis suggests PromptSpy currently exists as a proof of concept rather than a widespread operational threat. The examined samples appear to target users in Argentina specifically, with limited evidence of global distribution. Google's existing security infrastructure, including Play Protect, reportedly identifies and flags known variants of the malware.

This discovery builds upon ESET's previous research into AI-powered threats, including their identification of PromptLock ransomware for desktop systems in August 2025. The emergence of multiple AI-integrated threats within a relatively short timeframe indicates a deliberate trend among malware developers to explore artificial intelligence applications.

The implications for cybersecurity extend far beyond immediate threat mitigation. PromptSpy demonstrates that malicious actors are actively experimenting with AI integration, potentially leading to more adaptive and resilient threat variants. Traditional security approaches may prove insufficient against malware capable of real-time reasoning and adaptation.

For the artificial intelligence industry, PromptSpy represents a concerning milestone that highlights the dual-use nature of AI technology. As generative models become more accessible and powerful, the potential for malicious applications grows correspondingly. This case underscores the importance of implementing robust access controls and monitoring mechanisms for AI services.

The discovery also raises important questions about defensive strategies. Security teams may need to adopt AI-powered countermeasures to effectively combat adaptive threats. The traditional cat-and-mouse game between attackers and defenders is evolving into a more complex landscape where both sides leverage artificial intelligence capabilities.

Looking forward, PromptSpy likely represents the beginning of a new category of AI-enhanced threats rather than an isolated incident. As AI technology continues advancing and becoming more widely available, security professionals must prepare for increasingly sophisticated malware variants that can reason about their environment and adapt their behavior accordingly.

Enlaces citados