Full Report
Attackers are adapting established techniques to AI platforms, emerging technologies, and changing user behavior. ESET's new threat report examines the rise of malicious AI skills, AI-assisted malware, ClickFix attacks, record quishing activity, and ransomware tools designed to disable security software. [...]
Analysis Summary
# Tool/Technique: PromptSpy & AI-Assisted Malware
## Overview
PromptSpy is a specialized Android malware variant that integrates generative AI into its execution flow. Unlike traditional malware that uses hardcoded logic to interact with an operating system, this technique leverages Large Language Models (LLMs) to dynamically interpret and navigate user interfaces, allowing the malware to adapt to different device environments and security postures in real-time.
## Technical Details
- **Type:** Malware family (Android) / Technique (AI-Assisted execution)
- **Platform:** Android
- **Capabilities:** UI interpretation, dynamic execution flow, adaptive behavior.
- **First Seen:** Early 2026 (Reported by ESET)
## MITRE ATT&CK Mapping
- **[TA0040 - Impact]**
- **[T1499 - Endpoint Denial of Service]** (Via EDR Killer integration mentioned in the broader report context)
- **[TA0002 - Execution]**
- **[T1646 - Malicious AI Skills/Agents]** (Emerging sub-technique for AI-driven execution)
- **[TA0007 - Discovery]**
- **[T1418 - Software Discovery]** (Using AI to interpret UI elements and installed apps)
- **[TA0001 - Initial Access]**
- **[T1566.003 - Phishing: Spearphishing Link]** (Utilizing QR codes/Quishing)
## Functionality
### Core Capabilities
- **Generative AI Integration:** Utilizes Google’s Gemini AI to interpret mobile user interface (UI) elements.
- **Dynamic Adaptation:** Changes its behavior based on the specific layout and apps found on a target device, reducing the need for constant updates from a C2 server.
- **Social Engineering Expansion:** Leverages "ClickFix" techniques, using fake error messages and AI-themed help pages to trick users into granting permissions.
### Advanced Features
- **Heuristic Bypass:** By avoiding hardcoded strings and predictable execution patterns, it complicates traditional signature-based detection.
- **Contextual Awareness:** The malware can "understand" the context of a user's screen to determine the best moment to intercept data or inject malicious overlays.
## Indicators of Compromise
*Note: Specific hashes for the H1 2026 variants are contained within the full ESET technical report; general indicators based on the summary follow.*
- **File Hashes:** [Specific hashes not provided in summary text]
- **File Names:** Often disguised as "AI Assistant," "System Update," or "Browser Extension."
- **Network Indicators:**
- Communications with Gemini API endpoints (e.g., `generativelanguage.googleapis[.]com`).
- C2 traffic disguised as cloud authentication or AI help desk traffic.
- **Behavioral Indicators:** High frequency of Accessibility Service requests; unexpected automated interaction with UI elements without user input.
## Associated Threat Actors
- **General Cybercriminals:** Broad adoption by ransomware groups for EDR-killing.
- **Specialized Android Malware Operators:** Groups transitioning from traditional banking trojans to AI-assisted data theft.
## Detection Methods
- **Behavioral Detection:** Monitoring for the unauthorized use of Accessibility Services and unusual patterns of UI "scraping" or automated navigation.
- **API Monitoring:** Tracking anomalous or high-volume calls to LLM provider APIs (like Google Gemini) from non-standard mobile applications.
- **Network Traffic Analysis:** Identifying "ClickFix" domains and quishing-related redirects.
## Mitigation Strategies
- **Prevention:** Implement strict mobile device management (MDM) policies to restrict the installation of apps from unknown sources.
- **Hardening:** Disable "Accessibility Services" for all non-essential applications, as this is a primary vector for UI interpretation malware.
- **User Training:** Educate employees on "Quishing" (QR code phishing) and the "ClickFix" tactic where fake error messages prompt manual "fixes" that install malware.
## Related Tools/Techniques
- **ClickFix:** Social engineering technique using fake browser/service error messages.
- **Quishing:** The use of malicious QR codes to bypass email filters.
- **EDR Killers:** Specialized tools used by ransomware actors to disable security software prior to encryption.
- **AI Skills:** Small functional components used by malicious AI agents to automate tasks.