Course Objective & Scope
Penetration testers increasingly encounter fraud scoring models, image verification endpoints, recommendation engines and LLM-backed APIs alongside traditional web and REST services. This course equips offensive security professionals with a practical model penetration testing methodology spanning both predictive ML/DL models and generative AI services.
Following MITRE ATLAS tactics, participants perform black-box model reconnaissance, API fuzzing, evasion and extraction attacks with Counterfit and ART, and LLM exploitation with garak and PyRIT, culminating in an enterprise-grade model penetration test report.
Who Should Attend?
- Penetration testers and offensive security consultants
- Enterprise red team and application security (AppSec) teams
- Engineers conducting pre-release security assessments of AI services
- Prerequisite: experience with API penetration testing (Burp Suite / Python) and basic ML concepts
Key Takeaways
- Execute an end-to-end model penetration testing methodology grounded in MITRE ATLAS and OWASP
- Perform black-box reconnaissance of feature spaces, decision boundaries and underlying model families
- Run evasion, inversion and model extraction attacks against predictive APIs using Counterfit and ART
- Exploit hybrid API and prompt vulnerabilities in LLM endpoints with garak, PyRIT and Burp Suite
- Deliver a technical model pentest report complete with reproducible PoC scripts and remediation guidance
Curriculum
AI and Model Penetration Testing Methodology
- Combining traditional application/API pentesting with ML and LLM model testing
- Scoping, legal and data-privacy rules of engagement, and test harness setup
Model API Reconnaissance and Fingerprinting
- Inferring model families from input schemas, feature types, confidence scores and error messages
- Boundary probing, timing side-channels and semantic cache behaviour analysis
Penetration Testing Predictive (ML/DL) Models
- Black-box evasion, membership inference and surrogate extraction using Counterfit and IBM ART
- Testing file-upload endpoints (images, PDFs, audio, model artefacts) for parser and deserialisation flaws
Hybrid Penetration Testing of LLM and Agent APIs
- Automated probing with garak and PyRIT plus multi-step API + prompt manipulation via Burp Suite
- Pivoting into backend infrastructure through BOLA/IDOR, SSRF and unauthorised tool calling
Evidence Packaging, Risk Rating and Pentest Reporting
- Documenting reproducible PoC scripts, query cost metrics and business impact
- Writing layered remediation guidance for application, data science and platform teams
Other Courses in This Category
-
Hands-on training on applying machine learning and LLMs across cybersecurity—from phishing and malware detection to vulnerability prioritisation, cyber threat intelligence (CTI) analysis and SecOps assistants.
- Live Course
- 5 modules
-
Teaches how to detect malware, ransomware, C2 traffic and multi-stage attack chains across EDR, NDR and cloud telemetry using behavioural machine learning, sequence models and graph analytics.
- Live Course
- 5 modules
-
Teaches how to build unsupervised and semi-supervised anomaly detection pipelines on logs, network traffic and user behaviour (UEBA) using Isolation Forests, Autoencoders, time-series and graph methods while minimising false positives.
- Live Course
- 5 modules
-
Teaches how to integrate LLM assistants, natural-language-to-KQL/SPL query generation, automated alert enrichment, triage agents and incident summarisation safely into SIEM, SOAR and modern SOC operations.
- Live Course
- 5 modules
-
Covers model extraction, membership inference, model inversion, poisoning and evasion attacks against machine learning and deep learning models using the MITRE ATLAS framework, along with practical defenses.
- Live Course
- 5 modules
-
Hands-on training on FGSM, PGD, Carlini & Wagner, AutoAttack and black-box evasion attacks using the IBM Adversarial Robustness Toolbox (ART), paired with adversarial training and certified robustness defenses.
- Live Course
- 5 modules
Get in touch
