February 19, 2026 Source: BleepingComputer 3 min read · 600 words

PromptSpy is the first known Android malware to use generative AI at runtime

PromptSpy — перше відоме Android-шкідливе ПЗ, що використовує генеративний AI під час виконання

We've crossed a threshold. Security researchers just identified the first Android malware that doesn't just hide—it learns and adapts in real time using generative AI. PromptSpy, discovered and reported by BleepingComputer, represents something we haven't seen before: threat actors aren't just adopting AI as a tool anymore. They're embedding it into their malware as a core operational capability.

This isn't theoretical. This is happening right now on Android devices.

What We Know

According to BleepingComputer's report, PromptSpy s Google's Gemini model at runtime to dynamically adjust its persistence mechanisms across infected Android devices. Translation: the malware doesn't rely on a single, static technique to stay on your phone. It asks an AI system how to best hide itself on your specific device, then executes that strategy. Every infected device becomes a slightly different infection vector.

The timeline matters here. We're in February 2026, and this represents the first documented case of runtime AI integration in Android malware. Not a proof-of-concept. Not a lab demonstration. An actual campaign in the wild, targeting real devices.

What makes this particularly nasty because it collapses the window between detection and evolution.

How It Works

Traditional Android malware uses hardcoded persistence techniques—they latch onto system services, hide in app permissions, or exploit specific vulnerabilities. Once security vendors identify one technique, they can patch it across devices and flag it in detection signatures. The malware's playbook becomes static.

PromptSpy changes that equation entirely.

By querying Gemini at runtime, the malware essentially outsources its evasion strategy to a sophisticated AI model that can reason about device configurations, running processes, security posture, and detection patterns. It's not just hiding differently on each device—it's making intelligent decisions about how to hide based on what it observes about that specific environment. A device running aggressive security tools gets different persistence instructions than a minimally-protected one. The malware literally adapts to its surroundings.

And here's the compounding problem: this capability evolves faster than traditional threat intelligence cycles can track it.

Why It Matters

The android cyber attack landscape just entered a new era. Not because PromptSpy itself is the most dangerous malware ever created—it's probably not. But because it proves threat actors now understand how to weaponize AI for malware development at scale. Once one group does this, others will follow. Within months, we could see multiple android cyber security vulnerabilities emerge that specifically exploit AI-powered evasion techniques.

Android vulnerability reports from 2025 showed defenders struggling to keep pace with traditional exploits. Now add intelligent, self-optimizing persistence to that workload.

The real question is: how do you maintain detection signatures for malware that changes its behavior every time it runs?

Organizations relying on pattern-matching android cybersecurity tools are already at risk. Those tools identify threats by matching known malware signatures or behavioral patterns. Against PromptSpy, that's like hunting with a map from last year.

Next Steps

First: assume your android cyber security apps aren't sufficient if they're purely signature-based. Test your actual detection capabilities against adaptive malware—not just static samples. Demand visibility into behavioral analytics, not just file hashes.

Second: prioritize android vulnerability management immediately. Every unpatched CVE becomes an attack vector PromptSpy could use to establish initial access. Review your android vulnerability list from recent reports and treat high-severity items as blocking issues.

Third: implement endpoint detection and response (EDR) solutions designed for Android that can track process execution flows and AI model interactions, not just app installation. PromptSpy will hide itself, but its communication with Gemini leaves traces.

Finally: monitor your threat intelligence feeds religiously over the next 90 days. This malware campaign is new, and PromptSpy variants will emerge fast. What works today won't work next week.

The malware isn't smarter than you. But it's asking something that is.

Read original article →

Concerned about your project's security? Run an automated pentest with AISEC — AI-powered scanner with expert verification. Go to dashboard →