ABSTRACT
As cyber threats grow in volume and complexity, manual malware analysis can become a significant bottleneck for security teams. While AI promises acceleration, simple copy-paste interactions with web-based tools introduce privacy risks and fail to scale. This hands-on course teaches security professionals how to scale their analysis by building and deploying local, agentic workflows that automate reverse engineering tasks.
Moving beyond basic chat assistants, this training teaches students how to orchestrate agentic systems. Using Ghidra as the core static analysis engine, students will learn to integrate local Large Language Models (LLMs) directly into the analysis environment. The course covers advanced automation techniques, including scripting, headless Ghidra execution, custom loaders, and Model Context Protocol (MCP) server integration to allow agents to programmatically query and manipulate code.
A central theme is mastering the human-in-the-loop structures required to guide local agents. Students will learn to write focused prompts, develop custom agent tools, and audit LLM outputs. The course covers how to recognize and mitigate "model drift", where LLMs loop endlessly due to analysis complexity, and how to focus agents on relevant analysis tasks.
Through real-world case studies and hands-on labs, participants will advance from basic triage to reversing highly obfuscated malware. Students will perform binary triage, write custom binary loaders, and bypass evasion checks. Attendees will leave with the practical knowledge required to deploy a private, local agentic analysis stack in their own organization.
INTENDED AUDIENCE
- Malware analysts
- SOC analysts
- Incident responders
- Threat hunters
- Detection engineers
- Security practitioners
COURSE STRUCTURE
Malware Analysis Workflows Overview
Review the workflows related to malware analysis in increasing order of complexity. Malware analysis is only efficient and successful if the analyst has a clear goal and a solid understanding of the expected output.
- Triage: Determining if a set of files is malicious.
- IOC Extraction: Extracting related artifacts and network infrastructure.
- Detection Creation: Writing behavioral and static signatures to detect threat variants.
- Threat Tracking: Documenting and reimplementing algorithms to track malware campaigns over time.
Ghidra for Static Analysis
Understand core Ghidra capabilities to enhance or automate static analysis and prepare decompiler outputs for agent ingestion.
- Mastering disassembly, data, and decompilation views.
- Understanding memory, variables, and function prototypes to improve decompiler output.
- Navigating data structures and custom types.
- Scripting Ghidra to automate recurring analysis tasks.
Advanced Ghidra Features
Explore advanced Ghidra features that can be controlled by agents to accelerate reverse engineering.
- Emulating code execution from both the GUI and Python scripts.
- Developing custom Ghidra loaders to unpack, disassemble, and decompile payloads.
- Navigating the Debugger view for runtime validation.
- Automating and testing analysis workflows using Ghidra in headless mode.
- Setting up Model Context Protocol (MCP) servers to expose Ghidra's API to agents.
Scaling Binary Analysis with Local Models and Agentic Workflows
Deploy and configure local LLMs to accelerate and scale binary analysis using advanced agentic workflows.
- Designing targeted prompts to prevent agent hallucination and analysis drift.
- Developing custom agent skills to efficiently use tool capabilities.
- Implementing self-auditing workflows to double-check agent outputs for errors.
Dynamic Analysis and Agent Sandbox Orchestration
Leverage agents to orchestrate dynamic execution environments and focus static analysis on active code paths.
- Evaluating sandbox limitations and identifying VM fingerprinting attempts.
- Building agents to parse network traffic (PCAP) and extract dynamic indicators.
- Exposing execution traces to agents to identify relevant code areas in Ghidra.
Adversarial AI and Pipeline Resilience
Examine how adversaries target AI-assisted analysis pipelines and build resilient defensive workflows.
- Analyzing prompt injection risks within binary strings and headers.
- Studying real-world instances of adversarial prompt manipulation in malware.
- Building guardrails to detect anti-analysis and evasion tricks.
Prerequisites
- Programming & Scripting (Intermediate): Ability to read and write Python scripts and read C/C++ code
- Reverse Engineering & Debugging (Intermediate): Hands-on experience with static analysis tools, and basic debugging workflows.
- OS Internals & Architecture (Basic to Intermediate): Familiarity with Windows operating system internals.
Hardware and Software Requirements
- Hardware: Laptop with a multi-core CPU, minimum 16 GB RAM (32 GB strongly recommended to run local LLMs and debuggers simultaneously), and 50 GB of free disk space.
- Operating System: Windows 10/11 (with WSL2 enabled), macOS, or Linux.
- Software Prerequisites:
- Ghidra (latest stable release) with Java Development Kit (JDK 17 or 21).
- Python 3.10+ installed with a package manager (e.g.,
pip). - Ollama (or equivalent local model runner) for hosting 8B parameter models.
- A virtualization platform (Docker Desktop, VirtualBox, or VMware Workstation).
- A modern text editor or IDE (such as VS Code).
- All laboratory files and custom agent code templates will be provided in class.
YOUR INSTRUCTOR: Pierre-Marc Bureau
Pierre-Marc Bureau is an independent security researcher. He has more than 20 years of experience in malware analysis, threat intelligence, reverse engineering, and the disruption of large-scale criminal operations. Over the last decade, he has held several roles at Google — first on Chrome, then on Safe Browsing, and most recently within the Threat Analysis Group (now part of Google's Threat Intelligence Group). Across the roles, he has focused on protecting billions of users from malware and phishing. He has also supported external partners and internal Google teams in combatting financially motivated threat actors.
Before joining Google, he worked at ESET and Dell SecureWorks. At both ESET and Google, he has built and led teams of world-class threat analysts. He has designed multiple CTF challenges for the NorthSec competition and has presented at international conferences including Black Hat Europe, Recon, Hack.lu, and Virus Bulletin.
Cancellations are not permitted but attendee changes can be accommodated anytime prior to the start of the course.
Note: In the event of a class cancellation, Ringzer0 will endeavor to offer transfer to another training at no additional charge.
