Automating Malware Reverse Engineering with Local LLMs, PyGhidra and Neo4j Graphs
A researcher has built an automated system that combines PyGhidra, Neo4j and a local large language model to speed up the initial analysis of malicious binaries. The project addresses the main drawbacks of sending decompiled code to cloud LLMs: leakage of indicators of compromise, possible censorship of sensitive modules, and rapid context overflow when a sample contains 100–200 functions.
Architecture and data flow
The pipeline begins with ghidra_extractor.py, which uses PyGhidra to disassemble and decompile a binary without launching the Ghidra GUI. For every function the extractor records its address, name, size, cyclomatic complexity, decompiled C pseudocode (truncated at 6000 characters with a 30-second timeout), call-graph neighbors, string references and imported API calls. Metadata such as PE format, architecture and all standard file hashes are also stored. All extracted entities are loaded into Neo4j as distinct node types: Sample, Function, String, Import, Analysis, Capability and Behavior.
Next, worker.py and mcp.py iterate over functions and send each one to a locally hosted Qwen3 model running inside LM Studio. The prompt includes both the decompiled code and the function’s position in the call graph, allowing the model to understand caller-callee relationships. The model returns structured JSON describing the function’s purpose, extracted IOCs, tags, network indicators, command-line usage and anti-analysis techniques.
aggregator.py then aggregates tags across all functions into Capability nodes that carry a strength weight reflecting how often each capability appears. builder.py combines these capabilities with the call graph to detect higher-level behavioral patterns such as file encryption combined with network activity. Finally, generator.py produces a human-readable report by querying the enriched graph.
Graph data model
Storing functions in a flat table makes it difficult to answer questions such as whether five anti-debug functions call one another. In the Neo4j model a Function node is connected to other Function nodes via CALLS edges, to String nodes via REFERENCES edges, and to Import nodes via USES edges. Capability and Behavior nodes sit above the function level and are linked through weighted relationships, enabling concise Cypher queries instead of recursive joins.
Validation on WannaCry
The pipeline was tested on a WannaCry sample obtained from MalwareBazaar. Processing 195 functions took 126 minutes on a machine with 60 GB RAM and an Nvidia 5070 Ti. The generated report identified three behavioral patterns with 100 percent confidence: File Encryption (supported by 40 functions), C2 Communication and Anti-Analysis. A publicly available YARA rule confirmed the anti-debug findings discovered by the system.
The complete source code is available in the repository Dvoranchik/MLMalwareAnalyzer. Future work will focus on refining call-graph traversal and prompt engineering to surface additional indicators.
Related articles
MegaFon Reports 34% Rise in Russian Mobile Malware Infections Driven by Android Trojans
MegaFon monitoring data shows a 34% year-over-year increase in Russian users with detected mobile malware during summer 2026, with the sharpest spike of 77% recorded in August. The growth is primarily attributed to Android trojans capable of installing apps without user consent, intercepting notifications, enabling remote access, and stealing personal data. At the same time, blocked suspicious calls dropped significantly, by 31% in June, 41% in July, and 75% in August, which the operator links to state and corporate anti-fraud measures. Scammers are adapting by hijacking messenger accounts through social engineering, malware, and even rental schemes, often combining calls with fake app installations that deploy spyware. Phishing remains active, with 30% of detected malicious resources tied to investments and crypto, 21% to social networks and messengers, and 9% to harmful APK files. Director Sergey Khrenov noted that smartphones are increasingly used as tools by the attackers themselves in these hybrid schemes.
Reflective DLL Injection: How Malware Executes DLLs Directly in Memory Without Disk Files
Reflective DLL Injection allows malicious code to load and execute a DLL entirely in memory, bypassing traditional file-based loading mechanisms on Windows. The technique uses a custom loader inside the DLL itself to handle PE format parsing, relocations, imports, TLS, and exception handling without calling LoadLibrary. Unlike standard loading, the resulting module often resides in MEM_PRIVATE regions and does not appear in conventional module lists, complicating forensic analysis. Tools such as Meterpreter and Cobalt Strike have historically employed reflective loaders, though Beacon Object Files represent a different COFF-based approach. Investigators must examine memory regions, thread start addresses, and Sysmon events 7, 8, and 10 rather than relying solely on file artifacts. The method does not bypass DEP, Arbitrary Code Guard, or access-control restrictions and leaves detectable traces when combined with behavioral indicators. Proper detection requires correlating private executable memory with process behavior and avoiding over-reliance on simplified signatures.
Leaked DarkSword iOS Exploit Chain 'P7' Now Steals Crypto Wallet Seeds and Keystores
A third build of the P7 variant of the leaked DarkSword iOS exploit chain has been identified, featuring a new lure site themed around Chinese online casinos and a fresh command server. The chain exploits six vulnerabilities, including three zero-days, to deploy an implant that decrypts the keychain directly on the device and extracts seed phrases from wallets such as imToken, Trust Wallet, and Phantom. Unlike the original GHOSTBLADE, P7 rewrites the C2 agent to focus exclusively on wallet data while retaining the core TaskRop and MIG-filter bypass modules. Passive analysis of public sources also uncovered the long-running 'qqtime' delivery cluster that pairs DarkSword with the older Coruna chain for broader iOS coverage. The operator uses channel codes to support multiple affiliate distributors and employs an AppleKeyStore oracle to decrypt keychain items locally before exfiltration.
Censys Exposes DarkSword iOS Exploit Platform and Coruna Crypto Wallet Stealer
Censys has disclosed the inner workings of DarkSword, a commercial platform that sells remote access to iOS devices, along with its associated malware Coruna that targets cryptocurrency wallet recovery keys. The infrastructure was exposed between September 15 and 17, allowing researchers to analyze the full attack chain starting from a WebKit and JavaScriptCore exploit delivered through the browser. After escaping the Safari sandbox and reaching the kernel, the platform deploys three layers including a flag, controller, and main implant. Coruna then scans the device for BIP39-compliant seed phrases stored in photos and Apple Notes across 19 targeted wallet applications such as MetaMask, Phantom, Trust Wallet, Coinbase, Exodus, imToken, Bitpie, and BitKeep. The exposed server contained 11 victim recovery keys, 179 directories of extracted data, and 75 operator accounts, indicating a structured commercial operation with agent accounts, commissions, and device quotas. Confirmed infections affect devices running iOS 16.1 and 16.3.1, while Apple has extended patches to additional iOS 18 devices. Maintaining updated iOS versions remains the primary defense against this threat.