Loading story…
Researchers reported that several widely used Android malware detectors are misclassifying legitimate high-permission apps as malicious because they rely heavily on static signals such as dangerous permissions, sensitive API calls, and call-graph features without judging behavior in context. In a benign test set of 270 Android apps from AndroZoo that each requested at least nine dangerous permissions, tools including Drebin, MalScan, and MaskDroid reportedly flagged more than half as malware, while LAMD flagged about 80%, underscoring how privileged but legitimate apps can be over-identified as threats.
The studies highlight PRAXIS, a three-stage, LLM-assisted framework from researchers at Singapore Management University and Nankai University that combines manifest, bytecode, native-library, reachability, UI, and contextual analysis to distinguish suspicious behavior from legitimate app functionality. PRAXIS uses hypothesis generation, execution-chain confirmation, and contextual judgment tied to user awareness and app-role alignment; it reportedly reduced false positives to about 13% on the benign set and missed roughly one malware sample in nine when all stages were enabled, though researchers noted limits including cost, attacker-controlled contextual signals, and possible LLM training-data contamination.
Track how attackers are adapting to this technology.
4 events from the most recent confirmed update back to the earliest known activity.
In reported evaluations on Android samples from 2024 to 2026, PRAXIS achieved better detection performance and lower false-positive rates than traditional signal-based and isolated LLM-based scanners. One reported result said PRAXIS flagged 13% of the benign set and missed about one malware sample in nine when all stages were used.
Researchers introduced PRAXIS, a three-stage LLM-assisted Android malware detection framework that combines hypothesis generation, execution-chain confirmation, and contextual judgment using signals such as manifest, bytecode, native libraries, reachability, UI, and app-role alignment.
Using the benign high-permission app set, the researchers found that six Android malware detectors, including Drebin, MalScan, and MaskDroid, flagged more than half of the apps as malicious, while LAMD flagged 80%. The results highlighted high false-positive rates when detectors lacked contextual analysis.
Researchers from Singapore Management University and Nankai University assembled a benign test set of 270 Android apps from AndroZoo, each declaring at least nine dangerous permissions, to evaluate malware detector false positives.
Vulnerabilities, threat actors, malware, products, organizations, and breaches Mallory has linked to this story.
Follow how adversaries are adapting to this technology, and where it touches your stack today.
2 references tracked. Mallory keeps watching after this page renders.
cybersecuritynews.com
Open sourcehelpnetsecurity.com
Open sourceMap indicators from this story to your assets and identify affected systems in minutes.
Every observed campaign, victim, and pivot linked to actors named in this story.
Malware, exploits, and IOCs connected to the activity described here.
YARA, Sigma, and Snort rules deployed to your SIEM as soon as they’re published.
Get matching new stories delivered to your team as they break — not the next morning.
Ask questions about this story and take action on the answers.
Other stories Mallory is tracking that overlap on entities, sources, or topic labels.
Multiple reports highlight evolving **Android** threats that abuse SMS/telephony access and advanced evasion to enable fraud, surveillance, and account takeover. CloudSEK described a shift from repackaged apps to **runtime manipulation** using the *LSPosed* framework, where a malicious module (e.g., **Digital Lutera**) hooks `SmsManager` and `TelephonyManager` to undermine India’s **UPI SIM-binding** controls. The technique can intercept registration tokens and 2FA, spoof device identity/phone number, and exfiltrate data to **Telegram**; it also uses **Socket.IO** for real-time C2 and can remotely inject fabricated SMS entries into the device’s “Sent” database to make bank backends believe a SIM is present on a different device, enabling scalable payment fraud and account takeover. Separately, Acronis TRU (reported by Hackread) identified a **fake Red Alert** rocket-warning app distributed via SMS lures impersonating Israel’s Home Front Command; the trojanized app displays legitimate alerts to reduce suspicion while requesting extensive permissions to steal **GPS location**, **SMS/OTP**, contacts, installed-app inventory, and on-device account details, then exfiltrates data to a remote server, including via **certificate spoofing** and UI tricks to appear Play Store-installed. Zimperium reported a new Android RAT, **SurxRAT**, that can download and run **LLM modules** from third-party repositories to automate phishing and social engineering and to interact with apps/UI for credential theft and data exfiltration, reinforcing the need for behavior-based mobile detection, tighter app controls, and stronger integrity enforcement (e.g., *Play Integrity API* with `MEETS_STRONG_INTEGRITY`) where applicable.
Jun 29, 2026
Reporting highlighted ongoing **Android malware** activity spanning banking trojans, spyware, backdoors, and data stealers, with campaigns relying on social engineering, fake apps, stealthy persistence, and evasion techniques that can bypass signature-based defenses. The write-up emphasizes that mobile threats increasingly use dynamic payload delivery and credential/financial data theft, reinforcing the need for stronger mobile security controls such as behavior-based monitoring, careful app sourcing, least-privilege permissions, and timely patching. Separate analysis described the rise of **AI-driven scraping against mobile apps**, where attackers target the *mobile API surface* rather than the UI, leveraging the structured nature of API responses (e.g., predictable endpoints and clean JSON). The post outlines a common workflow—downloading an Android APK, reverse engineering it to extract endpoints and auth flows, and using runtime instrumentation in emulators or rooted devices with tools such as `JADX`, `Frida`, `Ghidra`, and `APKTool`—to automate data extraction at scale, highlighting that server-side systems often over-trust the client app and that traditional web anti-bot controls may not translate cleanly to mobile.
Jun 29, 2026
Google and Mandiant disclosed **PROMPTSPY**, an Android backdoor that uses AI-enabled functions to support credential theft, persistence, and operator resilience. The malware can capture biometric authentication input, display deceptive overlays to hinder removal, and relaunch remotely through Firebase Cloud Messaging. Investigators said operators can update command-and-control infrastructure, Gemini API keys, and VNC relay servers without redeploying the payload, allowing the malware to adapt after installation. Google said it disabled infrastructure tied to the activity, found no PROMPTSPY-infected apps on Google Play, and confirmed that Google Play Protect detects known samples. The report also described a broader rise in adversarial use of large language models and agentic tooling across the intrusion lifecycle. Google linked a suspected **PRC-nexus** actor to the use of tools including `Hexstrike`, `Graphiti`, `Strix`, `subfinder`, and `httpx` while targeting a Japanese technology company and an East Asian cybersecurity platform. Separately, the company said information operation actors tied to Russia, Iran, China, and Saudi Arabia are using AI for content generation, localization, reconnaissance, phishing preparation, environmental fingerprinting, and more autonomous attack workflows, including suspected AI voice cloning in the pro-Russia **Operation Overload** campaign.
Jun 29, 2026