GIAC Offensive AI Analyst

Testers need to know what the tools underneath actually do. You learn natural language processing, generative methods including GANs, retrieval-augmented generation with vector databases, custom assistants and the main AI platforms.

• Natural language processing in plain terms
• Generative models: language, image and audio
• GANs and how synthetic media is produced
• RAG and vector databases
• Custom assistants and their configuration
• The main AI platforms and their differences

You build A comparison of three AI platforms by capability, data handling and suitability for authorized testing.

Technology is rarely the first target. You examine the attack surfaces, objectives and psychology of social engineering in penetration tests, and the legal and ethical considerations that keep an assessment safe for the people in it.

• Social engineering attack surfaces
• Objectives of a social engineering assessment
• The psychology of persuasion and urgency
• Pretexts and how they are chosen for a test
• Legal and ethical considerations for staff testing
• Measuring results without blaming individuals

You build A social engineering assessment plan with objectives, safeguards for staff and success measures.

Reconnaissance sets up every later phase. You apply AI to sort and summarize public information during a test, use common data-gathering tools, and keep the work within ethical and legal limits.

• Reconnaissance in the penetration testing process
• Common OSINT data-gathering tools
• AI for summarizing and linking public information
• Verifying what AI reports before relying on it
• Safe and ethical limits on personal data
• Turning findings into a recon report

You build A reconnaissance report on a practice organization that uses AI summarization with every claim checked.

Phishing remains the most common way in, and AI makes convincing messages cheap. You use campaign tools and prompt design to build authorized simulations, measure the response, and give the client what it needs to improve its defenses.

• Anatomy of a phishing message and why it works
• Campaign tools such as GoPhish and the Social Engineering Toolkit
• Prompt design for realistic simulated messages
• Authorization, safeguards and staff communication
• Measuring clicks, reports and credential entry
• Defenses: filtering, training and reporting paths

You build An authorized phishing simulation plan, with message design notes, safeguards and a results report.

Synthetic audio, images and video give social engineering new credibility. You learn the tools and techniques behind deepfakes, what makes them persuasive, and the controls and checks an organization can adopt against them.

• How audio, image and video deepfakes are made
• Tools used and the effort each requires
• Why synthetic media persuades people
• Detection cues and detection tools
• Verification procedures for voice and video requests
• Including deepfake scenarios in authorized assessments

You build A verification procedure for high-risk voice and video requests, with an authorized test scenario to check it.

AI-assisted tools make more sense when you know what malware is built from. You review payloads, exploits and propagation, droppers and downloaders, and the common persistence, anti-detection and anti-analysis mechanisms.

• Payloads, exploits and propagation vectors
• Droppers and downloaders
• Persistence mechanisms at a conceptual level
• Anti-detection and anti-analysis ideas
• How malware is categorized and named
• What defenders look for in each stage

You build A malware component diagram for one sample family, with the detection point at each stage.

Models can write code that behaves like malware, which matters for testers and for defenders. You learn what authorized testing can use, how safety guardrails limit models and why testers must record when they meet them, and how AI-written samples are controlled and detected.

• What AI models can and cannot produce reliably
• Authorization needed before generating any sample
• Why model guardrails exist and how testers document them
• Safe handling of generated samples in a lab
• Reviewing AI-written code for errors and unsafe behavior
• Detecting and analyzing AI-generated malicious code

You build A control procedure for any AI-generated test sample, from authorization through lab handling to disposal.

Controls are only valuable if they hold under test. You learn how the main defensive controls on Windows are built, including the architecture of Windows Defender, where the design leaves openings, and how AI changes the effort needed to test them.

• How endpoint defenses detect and block
• Windows Defender architecture in outline
• Where design choices leave gaps for testers
• AI models as an aid in control testing
• Reporting control gaps with the fix
• Retesting after a defense is improved

You build A defensive control test plan for one endpoint configuration, with expected detections and a gap report.

Scan data is plentiful and context is scarce. You perform network enumeration and vulnerability scanning, then feed results into AI agent workflows that help find and explain weaknesses, with checks to catch the agent’s mistakes.

• Network enumeration and vulnerability scanning methods
• Choosing and configuring scanning tools
• Structuring scan data for an AI agent
• Contextualizing findings with agent assistance
• Validating agent conclusions before reporting
• Limits on agent actions during live testing

You build A scan-to-findings workflow in which an AI agent proposes priorities and a tester verifies each one.

Web applications remain the most reachable target. You use AI-powered tools to speed up testing, understand SQL injection and command injection as vulnerability classes, confirm findings with care, and write defenses that remove the cause.

• How AI tools streamline web application testing
• SQL injection as a vulnerability class
• Command injection as a vulnerability class
• Confirming a finding without causing damage
• Where AI tools produce false positives and misses
• Developer-facing fixes and reporting

You build A web application test report for a lab application, with confirmed findings and developer fixes.

Apply every skill to one assessment. You plan the scope and approvals, combine reconnaissance, social engineering simulation and technical testing with AI assistance, and report what worked, what the controls caught and what to change.

• Scoping and approvals
• Reconnaissance and simulation planning
• Technical testing with AI assistance
• Controls, detection and gaps
• Reporting and recommendations

You build A complete authorized AI-assisted assessment report with findings, detection results and recommendations.